RFC 8802 | The Quality for Service (Q4S) Protocol | July 2020 |
Aranda, et al. | Informational | [Page] |
This memo describes an application-level protocol for the communication of end-to-end QoS compliance information based on the HyperText Transfer Protocol (HTTP) and the Session Description Protocol (SDP). The Quality for Service (Q4S) protocol provides a mechanism to negotiate and monitor latency, jitter, bandwidth, and packet loss, and to alert whenever one of the negotiated conditions is violated.¶
Implementation details on the actions to be triggered upon reception/detection of QoS alerts exchanged by the protocol are out of scope of this document; it is either application dependent (e.g., act to increase quality or reduce bit-rate) or network dependent (e.g., change connection's quality profile).¶
This protocol specification is the product of research conducted over a number of years; it is presented here as a permanent record and to offer a foundation for future similar work. It does not represent a standard protocol and does not have IETF consensus.¶
This document is not an Internet Standards Track specification; it is published for informational purposes.¶
This is a contribution to the RFC Series, independently of any other RFC stream. The RFC Editor has chosen to publish this document at its discretion and makes no statement about its value for implementation or deployment. Documents approved for publication by the RFC Editor are not candidates for any level of Internet Standard; see Section 2 of RFC 7841.¶
Information about the current status of this document, any errata, and how to provide feedback on it may be obtained at https://www.rfc-editor.org/info/rfc8802.¶
Copyright (c) 2020 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document.¶
The World Wide Web (WWW) is a distributed hypermedia system that has gained widespread acceptance among Internet users. Although WWW browsers support other, preexisting Internet application protocols, the primary protocol used between WWW clients and servers became the HyperText Transfer Protocol (HTTP) ([RFC7230], [RFC7231], [RFC7232], [RFC7233], [RFC7234], and [RFC7235]). Since then, HTTP over TLS (known as HTTPS and described in [RFC2818]) has become an imperative for providing secure and authenticated WWW access. The mechanisms described in this document are equally applicable to HTTP and HTTPS.¶
The ease of use of the Web has prompted its widespread employment as a client/server architecture for many applications. Many of such applications require the client and the server to be able to communicate with each other and exchange information with certain quality constraints.¶
Quality in communications at the application level consists of four measurable parameters:¶
Any other communication parameter, such as throughput, is not a network parameter because it depends on protocol window size and other implementation-dependent aspects.¶
The Q4S protocol provides a mechanism for quality monitoring based on an HTTP syntax and the Session Description Protocol (SDP) in order to be easily integrated in the WWW, but it may be used by any type of application, not only those based on HTTP. Quality requirements may be needed by any type of application that communicates using any kind of protocol, especially those with real-time constraints. Depending on the nature of each application, the constraints may be different, leading to different parameter thresholds that need to be met.¶
Q4S is an application-level client/server protocol that continuously measures session quality for a given flow (or set of flows), end-to-end (e2e) and in real time; raising alerts if quality parameters are below a given negotiated threshold and sending recoveries when quality parameters are restored. Q4S describes when these notifications, alerts, and recoveries need to be sent and the entity receiving them. The actions undertaken by the receiver of the alert are out of scope of the protocol.¶
Q4S is session-independent from the application flows to minimize the impact on them. To perform the measurements, two control flows are created on both communication paths (forward and reverse directions).¶
This protocol specification is the product of research conducted over a number of years and is presented here as a permanent record and to offer a foundation for future similar work. It does not represent a standard protocol and does not have IETF consensus.¶
The purpose of Q4S is to measure end-to-end network quality in real time. Q4S does not transport any application data. This means that Q4S is designed to be used jointly with other transport protocols such as Real-time Transport Protocol (RTP) [RFC3550], Transmission Control Protocol (TCP) [RFC0793], QUIC [QUIC], HTTP [RFC7230], etc.¶
Some existent transport protocols are focused on real-time media transport and certain connection metrics are available, which is the case of RTP and RTP Control Protocol (RTCP) [RFC3550]. Other protocols such as QUIC provide low connection latencies as well as advanced congestion control. These protocols transport data efficiently and provide a lot of functionalities. However, there are currently no other quality measurement protocols offering the same level of function as Q4S. See Section 1.4 for a discussion of the IETF's quality measurement protocols, One-Way Active Measurement Protocol (OWAMP) and Two-Way Active Measurement Protocol (TWAMP).¶
Q4S enables applications to become reactive under e2e network quality changes. To achieve it, an independent Q4S stack application must run in parallel with the target application. Then, Q4S metrics may be used to trigger actions on the target application, such as speed adaptation to latency in multiuser games, bitrate control at streaming services, intelligent commutation of delivery node at Content Delivery Networks, and whatever the target application allows.¶
Monitoring quality of service (QoS) in computer networks is useful for several reasons:¶
Monitoring enables e2e negotiation of QoS parameters, independently of the ISPs of both endpoints.¶
A protocol to monitor QoS must address the following issues:¶
The Quality for Service (Q4S) protocol is a message-oriented communication protocol that can be used in conjunction with any other application-level protocol. Q4S is a measurement protocol. Any action taken derived from its measurements are out of scope of the protocol. These actions depend on the application provider and may be application-level adaptive reactions, may involve requests to the ISP, or whatever the application provider decides.¶
The benefits in quality measurements provided by Q4S can be used by any type of application that uses any type of protocol for data transport. It provides a quality monitoring scheme for any communication that takes place between the client and the server, not only for the Q4S communication itself.¶
Q4S does not establish multimedia sessions, and it does not transport application data. It monitors the fulfillment of the quality requirements of the communication between the client and the server; therefore, it does not impose any restrictions on the type of application, protocol, or usage of the monitored quality connection.¶
Some applications may vary their quality requirements dynamically for any given quality parameter. Q4S is able to adapt to the changing application needs, modifying the parameter thresholds to the new values and monitoring the network quality according to the new quality constraints. It will raise alerts if the new constraints are violated.¶
The Q4S session lifetime is composed of four phases with different purposes: Handshake, Negotiation, Continuity, and Termination. Negotiation and Continuity phases perform network parameter measurements per a negotiated measurement procedure. Different measurement procedures could be used inside Q4S, although one default measurement mechanism is needed for compatibility reasons and is the one defined in this document. Basically, Q4S defines how to transport application quality requirements and measurement results between a client and server and how to provide monitoring and alerting, too.¶
Q4S must be executed just before starting a client-server application that needs a quality connection in terms of latency, jitter, bandwidth, and/or packet loss. Once the client and server have succeeded in establishing communication under quality constraints, the application can start, and Q4S continues measuring and alerting if necessary.¶
The quality parameters can be suggested by the client in the first message of the Handshake phase, but it is the server that accepts these parameter values or forces others. The server is in charge of deciding the final values of quality connection.¶
OWAMP [RFC4656] and TWAMP [RFC5357] are two protocols to measure network quality in terms of RTT, but they have a different goal than Q4S. The main difference is the scope: Q4S is designed to assist reactive applications, whereas OWAMP/TWAMP is designed to measure just network delay.¶
The differences can be summarized in the following points:¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
This section introduces the basic operation of Q4S using simple examples. This section is of a tutorial nature and does not contain any normative statements.¶
The first example shows the basic functions of Q4S: communication establishment between a client and a server, quality requirement negotiations for the requested application, application start and continuous quality parameter measurements, and finally communication termination.¶
The client triggers the establishment of the communication by requesting a specific service or application from the server. This first message must have a special URI [RFC3986], which may force the use of the Q4S protocol if it is implemented in a standard web browser. This message consists of a Q4S BEGIN method, which can optionally include a proposal for the communication quality requirements in an SDP body. This option gives the client a certain negotiation capacity about quality requirements, but it will be the server who finally decides the stated requirements.¶
This request is answered by the server with a Q4S 200 OK response letting the client know that it accepts the request. This response message must contain an SDP body with the following:¶
"alerting-mode" attribute: There are two different scenarios for sending alerts that trigger actions either on the network or in the application when measurements identify violated quality constraints. In both cases, alerts are triggered by the server.¶
Q4S-aware-network scenario: The network is Q4S aware and reacts by itself to these alerts. In this scenario, Q4S-ALERT messages are sent by the server to the client, and network elements inspect and process these alert messages. The alerting mode in this scenario is called Q4S-aware-network alerting mode.¶
Reactive scenario: As shown in Figure 1, the network is not Q4S aware. In this scenario, alert notifications are sent to a specific node, called an Actuator, which is in charge of making decisions regarding what actions to trigger: either to change application behavior to adapt it to network conditions and/or invoke a network policy server in order to reconfigure the network and request better quality for application flows.¶
The format of messages exchanged between the server stack and the Actuator doesn't follow Q4S codification rules; their format will be implementation dependent. In this way, we will call the messages sent from the server stack to the Actuator "notifications" (e.g., alert notifications) and the messages sent from the Actuator to the server stack in response to notifications "acknowledges" (e.g., alert acknowledges).¶
It is important to highlight that any Q4S 200 OK response sent by the server to the client at any time during the life of a quality session may contain an SDP body with new values of quality constraints required by the application. Depending on the phase and the state of the measurement procedure within the specific phase, the client will react accordingly to take into account the new quality constraints in the measurement procedure.¶
Once the communication has been established (i.e., the Handshake phase is finished), the protocol will verify that the communication path between the client and the server meets the quality constraints in both directions, from and to the server (Negotiation phase). This Negotiation phase requires taking measurements of the quality parameters: latencies, jitter, bandwidth, and packet loss. This phase is initiated with a client message containing a Q4S READY method, which will be answered by the server with a Q4S 200 OK response.¶
Negotiation measurements are achieved in two sequential stages:¶
Stage 0 measurements are taken through Q4S PING messages sent from both the client and the server. All Q4S PING requests will be answered by Q4S 200 OK messages to allow for bidirectional measurements.¶
Different client and server implementations may send a different number of PING messages for measuring, although at least 255 messages should be considered to perform the latency measurement. The Stage 0 measurements only may be considered ended when neither client nor server receive new PING messages after an implementation-dependent guard time. Only after Stage 0 has ended, can the client send a "READY 1" message.¶
After a pre-agreed number of measurements have been performed, determined by the measurement procedure sent by the server, three scenarios may be possible:¶
Stage 1 is optional. Its purpose is to measure the availability of application-needed bandwidth. If the "bandwidth" attribute is set to zero kbps in the SDP, the client can skip stage 1 by sending a "READY 2" message after completion of stage 0. Stage 1 measurements are achieved through Q4S BWIDTH messages sent from both the client and the server. Unlike PING messages, Q4S BWIDTH requests will not be answered.¶
If Stage 0 and 1 meet the application quality constraints, the application may start. Q4S will enter the Continuity phase by measuring the network quality parameters through the Q4S PING message exchange on both connection paths and raising alerts in case of violation.¶
Once the client wants to terminate the quality session, it sends a Q4S CANCEL message, which will be acknowledged by the server with another Q4S CANCEL message. Termination of quality sessions are always initiated by the client because Q4S TCP requests follow the client/server schema.¶
Figure 2 depicts the message exchange in a successful scenario.¶
Both client and server measurements are included in the PING and BWIDTH messages, allowing both sides of the communication channel to be aware of all measurements in both directions.¶
The following two examples show the behavior of the Q4S protocol when quality constraints are violated, and alerts are generated; and, later on, when the violation of quality constraints stops leading to the execution of the recovery process. The first example (Figure 3) shows the Q4S-aware-network alerting mode scenario:¶
In this Q4S-aware-network alerting mode scenario, the server may send Q4S alerts to the client at any time upon detection of violated quality constraints. This alerting exchange must not interrupt the continuity quality parameter measurements between client and server.¶
The second example depicted in Figure 4 represents the Reactive scenario, in which alert notifications are sent from the server stack to the Actuator, which is in charge of deciding to act over application behavior and/or to invoke a network policy server. The Actuator is an entity that has a defined set of different quality levels and decides how to act depending on the actions stated for each of these levels; it can take actions for making adjustments on the application, or it can send a request to the policy server for acting on the network. The policy server also has a defined set of different quality levels previously agreed upon between the Application Content Provider and the ISP. The Reactive alerting mode is the default mode.¶
At the end of any stage of the Negotiation phase, the server sends an alert notification to the Actuator if quality constraints are violated. During the period of time defined by the "alert-pause" attribute, no further alert notifications are sent, but measurements are not interrupted. This way, both the client and the server will detect network improvements as soon as possible. In a similar way during the Continuity phase, the server may send alert notifications at any time to the Actuator upon detection of violated quality constraints. This alerting exchange must not interrupt the continuity measurements between client and server.¶
Finally, in the Termination phase, Q4S CANCEL messages sent from the client to the server must be forwarded from the server to the Actuator in order to release possible assigned resources for the session.¶
Q4S is a text-based protocol and uses the UTF-8 charset [RFC3629]. A Q4S message is either a request or a response.¶
Both request and response messages use the basic format of Internet Message Format [RFC5322]. Both types of messages consist of a start-line, one or more header fields, an empty line indicating the end of the header fields, and an optional message-body. This document uses ABNF notation [RFC5234] for the definitions of the syntax of messages.¶
The start-line, each message-header line, and the empty line MUST be terminated by a carriage-return line-feed sequence (CRLF). Note that the empty line MUST be present even if the message-body is not.¶
generic-message = start-line CRLF *message-header CRLF CRLF [ message-body ] start-line = Request-Line / Status-Line¶
Much of Q4S's messages and header field syntax are identical to HTTP/1.1. However, Q4S is not an extension of HTTP.¶
Q4S requests are distinguished by having a Request-Line for a start-line. A Request-Line contains a method name, a Request-URI, and the protocol version separated by a single space (SP) character.¶
The Request-Line ends with CRLF. No CR or LF are allowed except in the end-of-line CRLF sequence. No linear whitespace (LWSP) is allowed in any of the elements.¶
Request-Line = Method SP Request-URI SP Q4S-Version CRLF¶
Q4S responses are distinguished from requests by having a Status-Line as their start-line. A Status-Line consists of the protocol version followed by a numeric Status-Code and its associated textual phrase, with each element separated by a single SP character. No CR or LF is allowed except in the final CRLF sequence.¶
Status-Line = Q4S-Version SP Status-Code SP Reason-Phrase CRLF¶
The Status-Code is a 3-digit integer result code that indicates the outcome of an attempt to understand and satisfy a request. The Reason-Phrase is intended to give a short textual description of the Status-Code. The Status-Code is intended for use by automata, whereas the Reason-Phrase is intended for the human user. A client is not required to examine or display the Reason-Phrase.¶
While this specification suggests specific wording for the Reason-Phrase, implementations MAY choose other text, for example, in the language indicated in the Accept-Language header field of the request.¶
The first digit of the Status-Code defines the class of response. The last two digits do not have any categorization role. For this reason, any response with a status code between 100 and 199 is referred to as a "1xx response", any response with a status code between 200 and 299 as a "2xx response", and so on. Q4S/1.0 allows following values for the first digit:¶
The status codes are the same as described in HTTP [RFC7231]. In the same way as HTTP, Q4S applications are not required to understand the meaning of all registered status codes, though such understanding is obviously desirable. However, applications MUST understand the class of any status code, as indicated by the first digit, and treat any unrecognized response as being equivalent to the x00 status code of that class.¶
The Q4S-ALERT, Q4S-RECOVERY, and CANCEL requests do not have to be responded to. However, after receiving a Q4S-ALERT, Q4S-RECOVERY, or CANCEL request, the server SHOULD send a Q4S-ALERT, Q4S-RECOVERY, or CANCEL request to the client.¶
Q4S header fields are identical to HTTP header fields in both syntax and semantics.¶
Some header fields only make sense in requests or responses. These are called request header fields and response header fields, respectively. If a header field appears in a message that does not match its category (such as a request header field in a response), it MUST be ignored.¶
These fields may appear in request and response messages.¶
In addition to HTTP header fields, these are the specific Q4S request header fields:¶
this header field contains a digital signature that can be used by the network, Actuator, or policy server to validate the SDP, preventing security attacks. The Signature is an optional header field generated by the server according to the pre-agreed security policies between the Application Content Provider and the ISP. For example, a hash algorithm and encryption method such as SHA256 [RFC6234] and RSA [RFC8017] based on the server certificate could be used. This certificate is supposed to be delivered by a Certification Authority (CA) or policy owner to the server. The signature is applied to the SDP body.¶
Signature= RSA ( SHA256 (<sdp>), <certificate> )¶
If the Signature header field is not present, other validation mechanisms MAY be implemented in order to provide assured quality with security and control.¶
this header field carries the measurements of the quality parameters in PING and BWIDTH requests. The format is:¶
Measurements: "l=" " "|[0..9999] ", j=" " "|[0..9999] ", pl=" " "|[0.00 .. 100.00] ", bw=" " "|[0..999999]¶
Where "l" stands for latency followed by the measured value (in milliseconds) or an empty space, "j" stands for jitter followed by the measured value (in milliseconds) or an empty space, "pl" stands for packet loss followed by the measured value (in percentage with two decimals) or an empty space, and "bw" stands for bandwidth followed by the measured value (in kbps) or an empty space.¶
its purpose is to provide a sanity check and allow the server to close inactive sessions. If the client does not send a new request before the expiration time, the server MAY close the session. The value MUST be an integer, and the measurement units are milliseconds.¶
In order to keep the session open, the server MUST send a Q4S alert before the session expiration (Expires header field), with the same quality levels and an alert cause of "keep-alive". The purpose of this alert is to avoid TCP sockets, which were opened with READY message, from being closed, specially in NAT scenarios.¶
Requests, including new requests defined in extensions to this specification, MAY contain message bodies unless otherwise noted. The interpretation of the body depends on the request method.¶
For response messages, the request method and the response status code determine the type and interpretation of any message body. All responses MAY include a body.¶
The Internet media type of the message body MUST be given by the Content-Type header field.¶
The body MUST NOT be compressed. This mechanism is valid for other protocols such as HTTP and SIP [RFC3261], but a compression/coding scheme will limit the way the request is parsed to certain logical implementations, thus making the protocol concept more implementation dependent. In addition, the bandwidth calculation may not be valid if compression is used. Therefore, the HTTP Accept-Encoding request header field cannot be used in Q4S with values different from "identity", and if it is present in a request, the server MUST ignore it. In addition, the response header field Content-Encoding is optional, but if present, the unique permitted value is "identity".¶
The body length in bytes MUST be provided by the Content-Length header field. The "chunked" transfer encoding of HTTP/1.1 MUST NOT be used for Q4S.¶
The Method token indicates the method to be performed on the resource identified by the Request-URI. The method is case sensitive.¶
Method = "BEGIN" | "READY" | "PING" | "BWIDTH" | "Q4S-ALERT" | "Q4S-RECOVERY" | "CANCEL" | extension-method extension-method = token¶
The list of methods allowed by a resource can be specified in an Allow header field [RFC7231]. The return code of the response always notifies the client when a method is currently allowed on a resource, since the set of allowed methods can change dynamically. Any server application SHOULD return the status code 405 (Method Not Allowed) if the method is known, but not allowed for the requested resource, and 501 (Not Implemented) if the method is unrecognized or not implemented by the server.¶
The BEGIN method requests information from a resource identified by a Q4S URI. The purpose of this method is to start the quality session.¶
This method is used only during the Handshake phase to retrieve the SDP containing the sess-id and all quality and operation parameters for the desired application to run.¶
When a BEGIN message is received by the server, any current quality session MUST be canceled, and a new session should be created.¶
The response to a Q4S BEGIN request is not cacheable.¶
The READY method is used to synchronize the starting time for the sending of PING and BWIDTH messages over UDP between clients and servers. Including the Stage header field in this method is mandatory.¶
This message is used only in Negotiation and Continuity phases, and only just before making a measurement. Otherwise (outside of this context), the server MUST ignore this method.¶
This message is used during the Negotiation and Continuity phases to measure the RTT and jitter of a session. The message MUST be sent only over UDP ports.¶
The fundamental difference between the PING and BWIDTH requests is reflected in the different measurements achieved with them. PING is a short message, and it MUST be answered in order to measure RTT and jitter, whereas BWIDTH is a long message and MUST NOT be answered.¶
PING is a request method that can be originated by either the client or the server. The client MUST also answer the server PING messages, assuming a "server role" for these messages during the measurement process.¶
Including the Measurements header field in this method is mandatory, and provides updated measurements values for latency, jitter, and packet loss to the counterpart.¶
This message is used only during the Negotiation phase to measure the bandwidth and packet loss of a session. The message MUST be sent only over UDP ports.¶
BWIDTH is a request method that can be originated by either the client or the server. Both client and server MUST NOT answer BWIDTH messages.¶
Including the Measurements header field in this method is mandatory and provides updated measurements values for bandwidth and packet loss to the counterpart.¶
This is the request message that Q4S generates when the measurements indicate that quality constraints are being violated. It is used during the Negotiation and Continuity phases.¶
This informative message indicates that the user experience is being degraded and includes the details of the problem (bandwidth, jitter, packet loss measurements). The Q4S-ALERT message does not contain any detail on the actions to be taken, which depend on the agreements between all involved parties.¶
Unless there is an error condition, an answer to a Q4S-ALERT request is optional and is formatted as a request Q4S-ALERT message. If there is an error condition, then a response message is sent. The response to a Q4S-ALERT request is not cacheable.¶
This method MUST be initiated by the server in both alerting modes. In the Q4S-aware-network alerting mode, the Q4S-ALERT messages are sent by the server to the client, advising the network to react by itself. In the Reactive alerting mode, alert notifications are triggered by the server stack and sent to the Actuator (see Figure 1, "Reactive Scenario").¶
Client----q4s----SERVER STACK--->ACTUATOR-->APP OR POLICY SERVER¶
The way in which the server stack notifies the Actuator is implementation dependent, and the communication between the Actuator and the network policy server is defined by the protocol and API that the policy server implements.¶
This is the request message that Q4S generates when the measurements indicate that quality constraints, which had been violated, have been fulfilled during a period of time ("recovery-pause"). It is used during the Negotiation and Continuity phases.¶
This informative message indicates that the "qos-level" could be increased gradually until the initial "qos-level" is recovered (the "qos-level" established at the beginning of the session that was decreased during violation of constraints. See Section 7.9). The Q4S-RECOVERY message does not contain any detail on the actions to be taken, which depends on the agreements between all involved parties.¶
The answer to a Q4S-RECOVERY request is formatted as a request Q4S-RECOVERY message. A Q4S-RECOVERY request MUST NOT be answered with a response message unless there is an error condition. The response to a Q4S-RECOVERY request is not cacheable.¶
Like the Q4S-ALERT message, the Q4S-RECOVERY method is always initiated by the server in both alerting modes. In the Q4S-aware-network alerting mode, the Q4S-RECOVERY messages are sent by the server to the client, advising the network to react by itself. In the Reactive alerting mode, recovery notifications are triggered by the server stack and sent to the Actuator (see Figure 1, "Reactive Scenario").¶
The purpose of the CANCEL message is the release of the Q4S Session-Id and the possible resources assigned to the session. This message could be triggered by the Q4S stack or by the application using the stack (through an implementation-dependent API).¶
In the same way as Q4S-ALERT, CANCEL must not be answered with a response message, but with an answer formatted as a request Q4S-CANCEL message.¶
In the Reactive scenario, the server stack MUST react to the Q4S CANCEL messages received from the client by forwarding a cancel notification to the Actuator, in order to release possible assigned resources for the session (at the application or at the policy server). The Actuator MUST answer the cancel notification with a cancel acknowledge towards the server stack, acknowledging the reception.¶
Q4S response codes are used for TCP and UDP. However, in UDP, only the response code 200 is used.¶
The receiver of an unknown response code must take a generic action for the received error group (1xx, 2xx, 3xx, 4xx, 5xx, 6xx). In case of an unknown error group, the expected action should be the same as with the 6xx error group.¶
This response indicates that the request has been received by the next-hop server and that some unspecified action is being taken on behalf of this request (for example, a database is being consulted). This response, like all other provisional responses, stops retransmissions of a Q4S-ALERT during the "alert-pause" time.¶
2xx responses give information about the success of a request.¶
3xx responses give information about the user's new location or about alternative services that might be able to satisfy the request.¶
The requesting client SHOULD retry the request at the new address(es) given by the Location header field.¶
4xx responses are definite failure responses from a particular server. The client SHOULD NOT retry the same request without modification (for example, adding appropriate header fields or SDP values). However, the same request to a different server might be successful.¶
The request could not be understood due to malformed syntax. The Reason-Phrase SHOULD identify the syntax problem in more detail, for example, "Missing Sequence-Number header field".¶
The server has definitive information that the user does not exist at the domain specified in the Request-URI. This status is also returned if the domain in the Request-URI does not match any of the domains handled by the recipient of the request.¶
The method specified in the Request-Line is understood, but not allowed for the address identified by the Request-URI.¶
The response MUST include an Allow header field containing a list of valid methods for the indicated address.¶
The resource identified by the request is only able to generate response entities that have content characteristics that are not acceptable according to the Accept header field sent in the request.¶
The server could not produce a response within a suitable amount of time, and the client MAY repeat the request without modifications at any later time.¶
The server is refusing to process a request because the request entity-body is larger than the one that the server is willing or able to process. The server MAY close the connection to prevent the client from continuing the request.¶
The server is refusing to process the request because the Request-URI is longer than the one that the server accepts.¶
The server is refusing to process the request because the message body of the request is in a format not supported by the server for the requested method. The server MUST return a list of acceptable formats using the Accept, Accept-Encoding, or Accept-Language header field, depending on the specific problem with the content.¶
The server cannot process the request because the scheme of the URI in the Request-URI is unknown to the server.¶
5xx responses are failure responses given when a server itself is having trouble.¶
The server encountered an unexpected condition that prevented it from fulfilling the request. The client MAY display the specific error condition and MAY retry the request after several seconds.¶
The server does not support the functionality required to fulfill the request. This is the appropriate response when a server does not recognize the request method, and it is not capable of supporting it for any user.¶
Note that a 405 (Method Not Allowed) is sent when the server recognizes the request method, but that method is not allowed or supported.¶
The server is temporarily unable to process the request due to a temporary overloading or maintenance of the server. The server MAY indicate when the client should retry the request in a Retry-After header field. If no Retry-After is given, the client MUST act as if it had received a 500 (Server Internal Error) response.¶
A client receiving a 503 (Service Unavailable) SHOULD attempt to forward the request to an alternate server. It SHOULD NOT forward any other requests to that server for the duration specified in the Retry-After header field, if present.¶
Servers MAY refuse the connection or drop the request instead of responding with 503 (Service Unavailable).¶
The server did not receive a timely response from an external server it accessed in attempting to process the request.¶
The server does not support, or refuses to support, the Q4S protocol version that was used in the request. The server is indicating that it is unable or unwilling to complete the request using the same major version as the client, other than with this error message.¶
In the case that the Q4S version is not supported, this error may be sent by the server in the Handshake phase just after receiving the first BEGIN message from client.¶
The server was unable to process the request because the message length exceeded its capabilities.¶
6xx responses indicate that a server has definitive information about a particular policy not satisfied for processing the request.¶
The Session-Id is not valid.¶
The "qos-level" requested is not allowed for the client/server pair.¶
The session is not allowed due to some policy (the number of sessions allowed for the server is exceeded, or the time band of the Q4S-ALERT is not allowed for the client/server pair, etc.).¶
The policy server does not authorize the Q4S-ALERT quality session improvement operation due to an internal or external reason.¶
This section describes the measurement procedures, the SDP structure of the Q4S messages, the different Q4S protocol phases, and the messages exchanged in them.¶
All elements of the IP network contribute to quality in terms of latency, jitter, bandwidth, and packet loss. All these elements have their own quality policies in terms of priorities, traffic mode, etc., and each element has its own way to manage the quality. The purpose of a quality connection is to establish end-to-end communication with enough quality for the application to function flawlessly.¶
To monitor quality constraints of the application, four phases are defined and can be seen in Figure 5:¶
The original goal of SDP was to announce necessary information for the participants and multicast MBONE (Multicast Backbone) applications. Right now, its use has been extended to the announcement and the negotiation of multimedia sessions. The purpose of Q4S is not to establish media stream sessions, but to monitor a quality connection. This connection may be later used to establish any type of session including media sessions; Q4S does not impose any conditions on the type of communication requiring quality parameters.¶
SDP will be used by Q4S to exchange quality constraints and will therefore always have all the media descriptions ("m=") set to zero.¶
The SDP embedded in the messages is the container of the quality parameters. As these may vary depending on the direction of the communication (to and from the client), all quality parameters need to specify the uplink and downlink values: <uplink> / <downlink> (see Section 7.5.3 for an example). When one or both of these values are empty, it MUST be understood as needing no constraint on that parameter and/or that direction.¶
The uplink direction MUST be considered as being the communication from the client to the server. The downlink direction MUST be considered as being the communication from the server to the client.¶
The SDP information can comprise all or some of the following parameters shown in the example below. This is an example of an SDP message used by Q4S included in the 200 OK response to a Q4S BEGIN request.¶
v=0 o=q4s-UA 53655765 2353687637 IN IP4 192.0.2.33 s=Q4S i=Q4S parameters t=0 0 a=qos-level:0/0 a=alerting-mode:Reactive a=alert-pause:5000 a=public-address:client IP4 198.51.100.51 a=public-address:server IP4 198.51.100.58 a=measurement:procedure default(50/50,75/75,5000,40/80,100/256) a=latency:40 a=jitter:10/10 a=bandwidth:20/6000 a=packetloss:0.50/0.50 a=flow:app clientListeningPort TCP/10000-20000 a=flow:app clientListeningPort UDP/15000-18000 a=flow:app serverListeningPort TCP/56000 a=flow:app serverListeningPort UDP/56000 a=flow:q4s clientListeningPort UDP/55000 a=flow:q4s clientListeningPort TCP/55001 a=flow:q4s serverListeningPort UDP/56000 a=flow:q4s serverListeningPort TCP/56001¶
As quality constraints may be changed by applications at any time during the Q4S session lifetime, any Q4S 200 OK response sent by the server to the client in the Negotiation and Continuity phases could also include an SDP body with the new quality requirements stated by the applications from then on. Therefore, in response to any PING request sent by the client to the server, the server could send a Q4S 200 OK with an embedded SDP message that specifies new quality constraints requested by the application.¶
The "qos-level" attribute contains the QoS level for uplink and downlink. Default values are 0 for both directions. The meaning of each level is out of scope of Q4S, but a higher level SHOULD correspond to a better service quality.¶
Appropriate attribute values: [0..9] "/" [0..9]¶
The "qos-level" attribute may be changed during the session lifetime, raising or lowering the value as necessary following the network measurements and the application needs.¶
The "alerting-mode" attribute specifies the player in charge of triggering Q4S alerts in the case of constraint violation. There are two possible values:¶
Appropriate attribute values: <"Q4S-aware-network"|"Reactive">¶
The "alerting-mode" attribute is optional, and if not present, Reactive alerting mode is assumed.¶
In the Q4S-aware-network scenario, the "alert-pause" attribute specifies the amount of time (in milliseconds) the server waits between consecutive Q4S-ALERT messages sent to the client. In the Reactive scenario, the "alert-pause" attribute specifies the amount of time (in milliseconds) the server stack waits between consecutive alert notifications sent to the Actuator. Measurements are not stopped in Negotiation or Continuity phases during this period of time, but no Q4S-ALERT messages or alert notifications are fired, even with violated quality constraints, allowing for either network reconfigurations or application adjustments.¶
Appropriate attribute values: [0..60000]¶
In the Q4S-aware-network scenario, the "recovery-pause" attribute specifies the amount of time (in milliseconds) the server waits for initiating the "qos-level" recovery process. Once the recovery process has started, the "recovery-pause" attribute also states the amount of time (in milliseconds) between consecutive Q4S-RECOVERY messages sent by the server to the client (in the Q4S-aware-network scenario) or between recovery notifications sent by the server stack to the Actuator (in the Reactive scenario).¶
Appropriate attribute values: [0..60000]¶
This attribute contains the public IP address of the client and the server. The server fills these attributes with its own public IP address and the public IP address of the first message received from the client in the Handshake phase.¶
The purpose of these attributes is to make available the addressing information to the network policy server or other external entities in charge of processing Q4S-ALERT messages.¶
Appropriate attribute values: <"client"|"server"> <"IP4"|"IP6"> <value of IP address>¶
The maximum latency (considered equal for uplink and downlink) tolerance is specified in the "latency" attribute, expressed in milliseconds. In the Q4S-aware-network scenario, if the latency constraints are not met, a Q4S-ALERT method will be sent to the client. In the Reactive scenario, if the latency constraints are not met, an alert notification will be sent to the Actuator. If the "latency" attribute is not present or has a 0 value, no latency constraints need to be met, and no measurements MAY be taken.¶
Appropriate attribute values: [0..9999]¶
The maximum uplink and downlink jitter tolerance is specified in the "jitter" attribute, expressed in milliseconds. In the Q4S-aware-network scenario, if the jitter constraints are not met, a Q4S-ALERT method will be sent to the client. In the Reactive scenario, if the latency constraints are not met, an alert notification will be sent to the Actuator. If the "jitter" attribute is not present or has a 0 value, no jitter constraints need to be met, and no measurements MAY be taken.¶
Appropriate attribute values: [0..9999] "/" [0..9999]¶
The minimum uplink and downlink bandwidth is specified in the "bandwidth" attribute, expressed in kbps. In the Q4S-aware-network scenario, if the bandwidth constraints are not met, a Q4S-ALERT method will be sent to the client. In the Reactive scenario, an alert notification will be sent to the Actuator. If the "bandwidth" attribute is not present or has a 0 value, no bandwidth constraints need to be met, and no measurements MAY be taken.¶
Appropriate attribute values: [0..99999] "/" [0..99999]¶
The maximum uplink and downlink packet loss tolerance is specified in the "packetloss" attribute expressed in percentage (two decimal accuracy). In the Q4S-aware-network scenario, if the packetloss constraints are not met, a Q4S-ALERT method will be sent to the client. In the Reactive scenario, an alert notification will be sent to the Actuator. If the "packetloss" attribute is not present or has a 0 value, no packet loss constraints need to be met, and no measurements MAY be taken.¶
Appropriate attribute values: [0.00 ..100.00] "/"[0.00 ..100.00]¶
These attributes specify the flows (protocol, destination IP/ports) of data over TCP and UDP ports to be used in uplink and downlink communications.¶
Several "flow" attributes can be defined. These flows identify the listening port (client or server), the protocol (TCP [RFC0793] or UDP [RFC0768]) with the range of ports that are going to be used by the application and, of course, by the Q4S protocol (for quality measurements). All defined flows ("app" and "q4s") will be considered within the same quality profile, which is determined by the "qos-level" attribute in each direction. This allows us to assume that measurements on "q4s" flows are the same as experienced by the application, which is using "app" flows.¶
During Negotiation and Continuity phases, the specified Q4S ports in the "flow:q4s" attributes of SDP will be used for Q4S messages.¶
The Q4S flows comprise two UDP flows and two TCP flows (one uplink and one downlink for each one), whereas application traffic MAY consist of many flows, depending on its nature. The Handshake phase takes place through the Q4S Contact URI, using the standard Q4S TCP port. However, the Negotiation and Continuity phases will take place on the Q4S ports (UDP and TCP) specified in the SDP.¶
The "clientListeningPort" is a port on which the client listens for server requests and MUST be used as the origin port of client responses. The "serverListeningPort" is a port on which the server is listening for incoming messages from the client. The origin port of server responses may be different than the "serverListeningPort" value.¶
If "clientListeningPort" is zero ("a=flow:q4s clientListeningPort TCP/0"), the client MAY choose one randomly per OS standard rules. Client ports inside the SDP must always be matched against actual received port values on the server side in order to deal with NAT/NAPT devices. If a zero value or incorrect value is present, the server must set the value to the received origin port in the next message with SDP (200 OK, ALERT, and CANCEL messages).¶
Attribute values: <"q4s"|"app"> <"serverListeningPort"|"clientListeningPort"> <"UDP"|"TCP"> <0..65535> [ "-" [0..65535]]¶
These attributes contain the measurement procedure and the results of the quality measurements.¶
Measurement parameters are included using the session attribute "measurement". The first measurement parameter is the procedure. Q4S provides a "default" procedure for measurements, but others like RTP/RTCP might be used and defined later. This document will only define and explain the "default" procedure.¶
In the initial client request, a set of measurement procedures can be sent to the server for negotiation. One measurement procedure line MUST be included in the SDP message for each proposed method. The server MUST answer with only one line with the chosen procedure.¶
For each procedure, a set of values of parameters separated by "," can be included in the same attribute line. The amount and type of parameters depends on the procedure type.¶
In the following example, the "default" procedure type is chosen:¶
a=measurement:procedure default(50/50,75/75,5000,40/80,100/256)¶
In the "default" procedure, the meaning of these parameters is the following:¶
There are four more "measurement" attributes:¶
a=measurement:latency 45 a=measurement:jitter 3/12 a=measurement:bandwidth 200/9800 a=measurement:packetloss 0.00/1.00¶
The "measurement:latency", "measurement:jitter", "measurement:bandwidth", and "measurement:packetloss" attributes contain the values measured for each of these quality parameters in uplink and downlink directions. Notice that latency is considered equal for uplink and downlink directions. Quality parameter values in these "measurement" attributes provide a snapshot of the quality reached and MUST only be included in Q4S-ALERT messages in the SDP body such that they can be protected from malicious attacks as these alerts include a signature of the SDP body in the header. The rest of the messages will include the measured values in the Measurements header field.¶
In the case of the "default" procedure, the valid values are as follows:¶
a=measurement:procedure default,[0..999]"/" [0..999] "," [0..999] "/" [0..999] "," [0..9999] "," [0..999]/[0..999] "," [0..999]/[0..999]¶
The adaptation of measurement traffic to approximate the actual data streams' characteristics is convenient to accurately estimate the expected QoS for applications. Particularly, packet size can have a remarkable effect on bandwidth estimations. Moreover, this can produce problems depending on the MTU of the end hosts and links along the path.¶
Therefore, the maximum content length MAY be set in an attribute denoted as "max-content-length". Its value MUST be given in bytes and MUST NOT include application, transport, network, or link layer headers, i.e., size of the content length at the application layer. If not set, the value MUST be 1000 bytes.¶
Furthermore, this attribute MAY be used to communicate MTU limits in endpoints, hence reducing possible bias as a result of network-layer fragmentation.¶
For instance:¶
a=max-content-length:1300¶
This section describes the way quality parameters are measured as defined by the "default" procedure. Measurements MUST be taken for any quality parameter with constraints, that is, specified in the SDP attributes with non-zero values. For absent attributes, measurements MAY be omitted.¶
Latency measurements will be performed if the "latency" attribute and/or the "a=measurement:latency" attribute are present and have non-zero values.¶
Q4S defines a PING method in order to exchange packets between the client and the server. Based on this PING exchange, the client and the server are able to calculate the round-trip time (RTT). The RTT is the sum of downlink latency (normally named "reverse latency") and uplink latency (normally named "forward latency").¶
At least 255 samples of RTT MUST be taken by the client and server. As the forward and reverse latencies are impossible to measure, the client and server will assume that both latencies are identical (symmetric network assumption). The latency will therefore be calculated as the statistical median value of all the RTT samples divided by 2.¶
Jitter measurements will be performed if the "jitter" attribute and/or the "a=measurement:jitter" attribute are present and have non-zero values.¶
The jitter can be calculated independently by the client and by the server. The downlink jitter is calculated by the client taking into account the time interval between PING requests as defined by the "measurement:procedure" attribute in the first or second parameter depending on the Q4S protocol phase. The client and the server MUST send these PING requests at the specified intervals. The client measures the downlink jitter, whereas the server measures the uplink jitter. Note that PING responses are not taken into account when calculating jitter values.¶
Every time a PING request is received by an endpoint (either server or client), the corresponding jitter value is updated with the statistical jitter value, which is the arithmetic mean of the absolute values of elapsed times calculated on the first 255 packets received.¶
Each endpoint sends a PING periodically with a fixed interval, and each value of "elapsed time" (ET) should be very close to this interval. If a PING message is lost, the ET value is doubled. Identifying lost PING messages, however, is not an issue because all PING messages are labeled with a Sequence-Number header field. Therefore, the receiver can discard this ET value.¶
In order to have the first jitter sample, the receiver MUST wait until it receives 3 PING requests, because each ET is the time between two PINGs, and a jitter measurement needs at least two ET.¶
The client measures the values of RTT and downlink jitter, and the server measures RTT and uplink jitter, but all measurements are shared with the counterpart by means of the Measurements header field of the PING message.¶
Bandwidth measurements will be performed if the "bandwidth" attribute and/or the "a=measurement:bandwidth" attribute is present and has non-zero values.¶
In order to measure the available bandwidth, both the client and the server MUST start sending BWIDTH messages simultaneously using the UDP control ports exchanged during the Handshake phase in the SDP message at the needed rate to verify the availability of the bandwidth constraint in each direction. The messages are sent during the period of time defined in the third parameter of the SDP "measurement:procedure default" attribute in milliseconds.¶
The goal of these measurements is not to identify the available bandwidth of the communication path, but to determine if the required bandwidth is available, meeting the application's constraints. Therefore, the requested bandwidth MUST be measured sending only the highest bitrate required by the bandwidth attribute. This is illustrated in Figure 6.¶
ALERTS are not expected during bandwidth measurement, but only at the end of the measurement time.¶
When measuring bandwidth, all BWIDTH requests sent MUST be 1 kilobyte in length (UDP payload length by default), they MUST include a Sequence-Number header field with a sequential number starting at 0, and their content MUST consist of randomly generated values to minimize the effect of compression elements along the path. The Sequence-Number MUST be incremented by 1 with each BWIDTH packet sent. If any measurement stage needs to be repeated, the sequence number MUST start at zero again. BWIDTH requests MUST NOT be answered. Examples:¶
Client message: ========================= BWIDTH q4s://www.example.com Q4S/1.0 User-Agent: q4s-ua-experimental-1.0 Session-Id: 53655765 Sequence-Number: 0 Content-Type: text Content-Length: XXXX Measurements: l=22, j=10, pl=0.00, bw=3000 VkZaU1FrNVZNVlZSV0doT1ZrZ (to complete up to "max-content- length" bytes UDP payload length) =========================¶
The client MUST send BWIDTH packets to the server to allow the server to measure the uplink bandwidth. The server MUST send BWIDTH packets to the client to allow the client to measure the downlink bandwidth.¶
Server message: ========================= BWIDTH q4s://www.example.com Q4S/1.0 Session-Id: 53655765 Sequence-Number: 0 Content-Type: text Content-Length: XXXX Measurements: l=22, j=7, pl=0.00, bw=200 ZY0VaT1ZURlZVVmhyUFE9PQ (to complete up to max-content- length UDP payload length) =========================¶
Packet loss and bandwidth are measured simultaneously using the BWIDTH packets sent by both the client and the server. Because the BWIDTH packets contain a Sequence-Number header field incremented sequentially with each sent packet, lost packets can be easily identified. The lost packets MUST be counted during the measurement time.¶
The first phase consists of a Q4S BEGIN method issued from the client to the server as shown in Figure 7.¶
The first Q4S message MUST have a special URI [RFC3986], which forces the use of the Q4S protocol if it is implemented in a standard web browser.¶
This URI, named "Contact URI", is used to request the start of a session. Its scheme MUST be:¶
"q4s:" "//" host [":" port] [path["?" query]¶
Optionally, the client can send the desired quality parameters enclosed in the body of the message as an SDP document. The server MAY take them into account when building the answer message with the final values in the SDP body, following a request/response schema [RFC3264].¶
If the request is accepted, the server MUST answer it with a Q4S 200 OK message, which MUST contain an SDP body [RFC4566] with the assigned sess-id (embedded in the SDP "o=" line), the IP addresses to be used, the flow ports to be used, the measurement procedure to be followed, and information about the required quality constraints. Additionally, the "alerting-mode" and "alert-pause" time attributes may be included. Q4S responses should use the protocol designator "Q4S/1.0".¶
After these two messages are exchanged, the first phase is completed. The quality parameter thresholds have been sent to the client. The next step is to measure the actual quality of the communication path between the client and the server and alert if the Service Level Agreement (SLA) is being violated.¶
The following is an example of a client request and a server answer:¶
Client Request: ========================= BEGIN q4s://www.example.com Q4S/1.0 Content-Type: application/sdp User-Agent: q4s-ua-experimental-1.0 Content-Length: 142 (SDP not shown) ========================= Server Answer: ========================= Q4S/1.0 200 OK Date: Mon, 10 Jun 2010 10:00:01 GMT Content-Type: application/sdp Expires: 3000 Signature: 6ec1ba40e2adf2d783de530ae254acd4f3477ac4 Content-Length: 131 (SDP not shown) =========================¶
The header fields used are explained in Section 4.3.¶
The Negotiation phase is in charge of measuring the quality parameters and verifying that the communication paths meet the required quality constraints in both directions as specified in the SDP body.¶
The measured parameters will be compared with the quality constraints specified in the SDP body. If the quality session is compliant with all the quality constraints, the application can start.¶
If the quality constraints are not met, a higher quality service level will be demanded. Depending on the scenario, this quality upgrade will be managed as follows:¶
In both scenarios stated above, if after several measurement cycles, the network constraints cannot be met, the quality session is terminated. Concretely when, under all possible actions taken by Actuator, the quality remains below requirements, the session must be terminated.¶
The steps to be taken in this phase depend on the measurement procedure exchanged during the Handshake phase. This document only describes the "default" procedure, but others can be used, like RTP/RTCP [RFC3550].¶
Measurements of latency and jitter are made by calculating the differences in the arrival times of packets and can be achieved with little bandwidth consumption. The bandwidth measurement, on the other hand, involves higher bandwidth consumption in both directions (uplink and downlink).¶
To avoid wasting unnecessary network resources, these two types of measurements will be performed in two separate stages. If the required latencies and jitters cannot be reached, it makes no sense to waste network resources measuring bandwidth. In addition, if achieving the required latency and jitter thresholds implies upgrading the quality session level, the chance of obtaining compliant bandwidth measurements without retries is higher, saving network traffic again. Therefore, the "default" procedure determines that the measurements are taken in two stages:¶
Notice that packet loss can be measured in both stages, as all messages exchanged include a Sequence-Number header field that allows for easy packet loss detection.¶
The client starts the Negotiation phase by sending a READY request using the TCP Q4S ports defined in the SDP. This READY request includes a Stage header field that indicates the measurement stage.¶
If either jitter, latency, or both are specified, the Negotiation phase begins with the measurement of latencies and jitters (stage 0). If none of those attributes is specified, stage 0 is skipped.¶
The Stage 0 MUST start with a synchronization message exchange initiated with the client's READY message.¶
Client Request, READY message: ========================= READY q4s://www.example.com Q4S/1.0 Stage: 0 Session-Id: 53655765 User-Agent: q4s-ua-experimental-1.0 Content-Length: 0 ========================= Server Response: ========================= Q4S/1.0 200 OK Session-Id: 53655765 Stage:0 Content-Length: 0 =========================¶
This triggers the exchange of a sequence of PING requests and responses that will lead to the calculation of RTT (latency), jitter, and packet loss.¶
After receiving a 200 OK, the client must send the first PING message, and the server will wait to send PINGs until the reception of this first client PING.¶
The client and server MUST send PING requests to each other. The Sequence-Number header field of the first PING MUST be set to 0. The client and server will manage their own sequence numbers.¶
The following is an example of the PING request sent from the client and the server's response:¶
Client Request: ========================= PING q4s://www.example.com Q4S/1.0 Session-Id: 53655765 Sequence-Number: 0 User-Agent: q4s-ua-experimental-1.0 Measurements: l=22, j=12, pl=0.20, bw= Content-Length: 0 ========================= Server Response: ========================= Q4S/1.0 200 OK Session-Id: 53655765 Sequence-Number: 0 Content-Length: 0 =========================¶
The function of the PING method is similar to the ICMP echo request message [RFC0792]. The server MUST answer as soon as it receives the message.¶
Both endpoints MUST send Q4S PING messages with the periodicity specified in the first parameter of SDP "measurement:procedure" attribute, always using the same UDP ports and incrementing the Sequence-Number with each message.¶
In the following example, the value of the first parameter of the SDP "measurement:procedure" attribute is 50 milliseconds (from the client to the server) and 60 ms (from the server to the client):¶
a=measurement:procedure default(50/60,50/50,5000,256/256,256/256)¶
They MUST NOT wait for a response to send the next PING request. The Sequence-Number header field value is incremented sequentially and MUST start at zero. If this stage is repeated, the initial Sequence-Number MUST start at zero again.¶
All PING requests MUST contain a Measurements header field with the values of the latency, jitter, and packet loss measured by each entity up to that moment. The client will send its measurements to the server, and the server will send its measurements to the client. Example:¶
Measurements: l=22, j=13, pl=0.10, bw=¶
Where "l" stands for latency, "j" for jitter, "pl" for packet loss, and "bw" for bandwidth. The bandwidth value is omitted, as it is not measured at this stage.¶
Optionally the PING request can include a Timestamp header field with the time in which the message has been sent. In the case that the header field is present, the server MUST include the header field in the response without changing the value.¶
A minimum number of PING messages MUST be exchanged in order to be able to measure latency, jitter, and packet loss with certain accuracy (at least 256 samples are RECOMMENDED to get an accurate packet loss measurement). Both the client and the server calculate the respective measured parameter values. The mechanisms to calculate the different parameters are described in Section 7.3.¶
At the end of this stage 0, there are three possibilities:¶
In the first case, Stage 0 is finished. The client and server are ready for Stage 1: bandwidth and packet loss measurement. The client moves to stage 1 by sending a READY message that includes the header field, "Stage: 1".¶
If the bandwidth constraints are either empty or have a value of zero, the Negotiation phase MUST terminate, and both client and server may initiate the Continuity phase. In this case, client moves to the Continuity phase by sending a READY message that includes the header field, "Stage: 2".¶
The second case, in which one or more quality constraints have not been met, is detailed in Section 7.5.4.¶
This stage begins in a similar way to stage 0, sending a READY request over TCP. The value of the READY message's Stage header field is 1. The server answers with a Q4S 200 OK message to synchronize the initiation of the measurements as shown in Figure 9.¶
Client Request: ========================= READY q4s://www.example.com Q4S/1.0 User-Agent: q4s-ua-experimental-1.0 Stage: 1 Session-Id: 53655765 Content-Length: 0 ========================= Server Response: ========================= Q4S/1.0 200 OK Session-Id: 53655765 Stage: 1 Content-Length: 0 =========================¶
Just after receiving the 200 OK, both the client and the server MUST start sending BWIDTH messages simultaneously using the UDP "q4s" ports. Section 7.3.3 describes the bandwidth measurement in detail.¶
At the end of this stage 1, there are three possibilities:¶
In the first case, Stage 1 is finished. The client and server are ready for the Continuity phase. The client moves to this phase by sending a READY message that includes the header field, "Stage: 2". The server answer MUST be 200 OK as shown in Figure 10.¶
Client Request: ========================= READY q4s://www.example.com Q4S/1.0 User-Agent: q4s-ua-experimental-1.0 Stage: 2 Session-Id: 53655765 Content-Length: 0 ========================= Server Answer: ========================= Q4S/1.0 200 OK Date: Mon, 10 Jun 2010 10:00:01 GMT Session-Id: 53655765 Trigger-URI: http://www.example.com/app_start Expires: 3000 Content-Type: application/sdp Signature: 6ec1ba40e2adf2d783de530ae254acd4f3477ac4 Content-Length: 131 (SDP not shown) =========================¶
If the Trigger-URI header field is present, the client SHOULD send an HTTP request to this URI.¶
The second case, with violated network constraints, is explained in Section 7.5.4.¶
After finishing Stage 1 of the Negotiation phase, the client and the server have each other's measured parameter values as these have been exchanged in the Measurements header fields of the PING and BWIDTH messages. If there is one or more parameters that do not comply with the uplink or downlink application constraints required, both the server and the client are aware of it.¶
If there is any quality parameter that does not meet the uplink or downlink quality constraints specified in the SDP message, two scenarios are possible depending on the specified alerting mode (if not present, the default value is Reactive alerting mode):¶
Q4S-aware-network alerting mode: the server MUST send a Q4S-ALERT message to the client including the digital Signature header field, and the client MUST answer with the same Q4S-ALERT message. The Signature header field contains the signed hash value of the SDP body in order to protect all the SDP data, and therefore it MUST contain the "measurement" parameters in the body.¶
Server request ========================= Q4S-ALERT q4s://www.example.com Q4S/1.0 Host: www.example.com User-Agent: q4s-ua-experimental-1.0 Session-Id: 53655765 Content-Type: application/sdp Content-Length: 142 v=0 o=q4s-UA 53655765 2353687637 IN IP4 192.0.2.33 s=Q4S i=Q4S parameters t=0 0 a=qos-level:1/2 a=alerting-mode: Q4S-aware-network a=alert-pause:5000 a=public-address:client IP4 198.51.100.51 a=public-address:server IP4 198.51.100.58 a=latency:40 a=jitter:10/10 a=bandwidth:20/6000 a=packetloss:0.50/0.50 a=flow:app downlink TCP/10000-20000 a=flow:app uplink TCP/56000 a=flow:q4s downlink UDP/55000 a=flow:q4s downlink TCP/55001 a=flow:q4s uplink UDP/56000 a=flow:q4s uplink TCP/56001 a=measurement:procedure default(50/50,50/50,5000,256/256,256/256) a=measurement:latency 30 a=measurement:jitter 6/4 a=measurement:bandwidth 200/4000 a=measurement:packetloss 0.20/0.33 =========================¶
At this point, both the client and server keep on measuring but without sending new Q4S-ALERT messages during the "alert-pause" milliseconds.¶
At this point during Negotiation phase, both the client and server keep on measuring without sending new alert notifications to the Actuator during the "alert-pause" milliseconds specified in the SDP. This way, both client and server will detect any improvement in network conditions as soon as the network reacts. The application can start as soon as the number of measurements indicated in the "measurement:procedure" attribute indicates that the quality parameters are met.¶
The same applies to Continuity phase: the measurement dialog between client and server must not be interrupted by any possible ALERT message.¶
The actuator receives notifications of unmet requirements from the Q4S server stack and acts upon the application or the network policy server, according to logic out of scope of this protocol.¶
The Actuator logic activates mechanisms at the application level and/or the network level based on a quality level dictionary, in which the meaning of each level is implementation dependent, and each level involves different actions based on rules to keep a certain user experience quality.¶
The type of actions that an Actuator can take at the application level are application dependent and MAY involve:¶
Apart from actions at the application level, the Actuator MAY act at the network level if a network policy server is available.¶
A network policy server may be part of the Reactive scenario, and it is in charge of managing network quality provision. A network policy server may implement all or some of these features (but implementation is not exclusive to):¶
Policy rules (the following rules are only examples):¶
If any of the policy rules fail, a Q4S-ALERT message MUST be answered by a 6xx error indicating the cause.¶
If any constraint was violated, the server MAY trigger a Q4S-ALERT asking for a higher "qos-level" attribute. The maximum "qos-level" allowed is 9 for both uplink and downlink.¶
If the "qos-level" has reached the maximum value for the downlink or uplink without matching the constraints, then a CANCEL request MUST be sent by the client using the TCP port determined in the Handshake phase in order to release the session. In reaction to the reception of the CANCEL request, the server MUST send a CANCEL request, too. If no CANCEL request is received, the expiration time cancels the session on the server side.¶
Client Request: ========================= CANCEL q4s://www.example.com Q4S/1.0 User-Agent: q4s-ua-experimental-1.0 Session-Id: 53655765 Content-Type: application/sdp Content-Length: 142 (SDP not shown) ========================= Server Request in reaction to Client Request: ========================= CANCEL q4s://www.example.com Q4S/1.0 Session-Id: 53655765 Expires: 0 Content-Type: application/sdp Signature: 6ec1ba40e2adf2d783de530ae254acd4f3477ac4 Content-Length: 131 (SDP not shown) =========================¶
During the Negotiation phase, latency, jitter, bandwidth, and packet loss have been measured. During the Continuity phase, bandwidth will not be measured again because bandwidth measurements may disturb application performance.¶
This phase is supposed to be executed at the same time as the real-time application is being used.¶
This document only covers the "default" procedure. The continuity operation with the "default" procedure is based on a sliding window of samples. The number of samples involved in the sliding window may be different for jitter and latency than for packet loss calculations according to the fifth and sixth parameters of the "measurement:procedure" attribute. In the example, shown in Figure 11, the jitter and latency sliding window comprises 40 samples, whereas the size of the packet loss sliding window is 100 samples:¶
a=measurement:procedure default(50/50,75/75,5000,40/40,100/100)¶
In addition, the sizes of these windows are configurable per direction: uplink and downlink values may differ.¶
PING requests are sent continuously (in both directions), and when the Sequence-Number header field reaches the maximum value, the client continues sending PING messages with the Sequence-Number header field starting again at zero. When the server PING Sequence-Number header field reaches the maximum value, it does the same, starting again from zero.¶
On the client side, the measured values of downlink jitter, downlink packet loss, and latency are calculated using the last samples, discarding older ones, in a sliding window schema.¶
Only if the server detects that the measured values (downlink or uplink jitter, packet loss, or latency) are not reaching the quality constraints, a Q4S-ALERT is triggered and sent either to the client or to the Actuator, depending on the alerting mode, and the "alert-pause" timer is started.¶
In the Q4S-aware-network alerting mode shown in Figure 12, if the client receives a Q4S-ALERT message, it MUST answer by sending the Q4S-ALERT request message including the SDP (with its corresponding digital signature) back to the server.¶
Both client and server will keep performing measurements, but Q4S-ALERT messages MUST NOT be sent during "alert-pause" milliseconds. The operations needed to act on the network and the agents in charge of them are out of scope of this document.¶
In the Reactive scenario shown in Figure 13, if the server detects that the measured values (downlink or uplink jitter, packet loss, or latency) are not reaching the quality constraints, an alert notification is triggered and sent to the Actuator. The Actuator MUST then answer to the server stack with an alert acknowledgement.¶
The measurement dialog between the client and the server MUST NOT be interrupted by any possible ALERT message.¶
The Termination phase is the endpoint for the established Q4S session that is reached in the following cases:¶
The meaning of the Termination phase in terms of the release of resources or accounting is application dependent and out of scope of the Q4S protocol.¶
In the Reactive alerting mode, Q4S CANCEL messages received by the Q4S server must cause the server stack to send cancel notifications to the Actuator in order to release possible assigned resources for the session.¶
A session may finish due to several reasons (client shutdown, client CANCEL request, constraints not reached, etc.), and any session finished MUST release the assigned resources.¶
In order to release the assigned server resources for the session, the Expires header field indicates the maximum interval of time without exchanging any Q4S message.¶
Depending on the nature of the application, the quality constraints to be reached may evolve, changing some or all quality constraint values in any direction.¶
The client MUST be able to deal with this possibility. When the server sends an SDP document attached to a response (200 OK or Q4S-ALERT, etc.), the client MUST take all the new received values, overriding any previous value in use.¶
The dynamic changes on the quality constraints can be a result of two possibilities:¶
TCP throughput can be less than actual bandwidth if the Bandwidth-Delay Product (BDP) is large, or if the network suffers from a high packet loss rate. In both cases, TCP congestion control algorithms may result in a suboptimal performance.¶
Different TCP congestion control implementations like Reno [RENO], High Speed TCP [RFC3649], CUBIC [CUBIC], Compound TCP (CTCP) [CTCP], etc., reach different throughputs under the same network conditions of RTT and packet loss. In all cases, depending on the RTT-measured value, the Q4S server could dynamically change the packetloss constraints (defined in the SDP) in order to make it possible to reach a required throughput or vice versa (using "measurement:packetloss" to change dynamically the latency constraints).¶
A general guideline for calculating the packet loss constraint and the RTT constraint consists of approximating the throughput by using a simplified formula, which should take into account the TCP stack implementation of the receiver, in addition to the RTT and packet loss:¶
Th= Function( RTT, packet loss, ...)¶
Then, depending on RTT-measured values, set dynamically the packet loss constraint.¶
It is possible to easily calculate a worst-case boundary for the Reno algorithm, which should ensure for all algorithms that the target throughput is actually achieved, except that high-speed algorithms will then have even larger throughput if more bandwidth is available.¶
For the Reno algorithm, the Mathis formula may be used [RENO] for the upper bound on the throughput:¶
Th <= (MSS/RTT)*(1 / sqrt{p})¶
In the absence of packet loss, a practical limit for the TCP throughput is the receiver_window_size divided by the RTT. However, if the TCP implementation uses a window scale option, this limit can reach the available bandwidth value.¶
Each time the server detects a violation of constraints, the alert mechanism is triggered, the "alert-pause" timer is started, and the "qos-level" is increased. When this happens repeatedly, and the "qos-level" reaches its maximum value (value 9), the session is canceled. But when the violation of constraints stops before reaching "qos-level" maximum value, the recovery mechanism allows for the "qos-level" upgrade gradually.¶
This downgrade and upgrade of "qos-level" is explained with the following example:¶
When the network configuration allows for the possibility of managing Q4S flows and application flows independently (either is a network-based QoS or a Q4S-aware network), the "qos-level" downgrade process could be managed more efficiently using a strategy that allows for carrying out "qos-level" downgrades excluding application flows from SDP dynamically. The Q4S flows would be downgraded to allow for measurements on a lower quality level without interference of the application flows. A Q4S client MUST allow this kind of SDP modification by the server.¶
Periodically (every several minutes, depending on the implementation) a Q4S-ALERT could be triggered, in which the level is downgraded for Q4S flows, excluding application flows from the embedded SDP of that request.¶
This mechanism allows the measurement at lower levels of quality while application flows continue using a higher "qos-level" value.¶
This mechanism, illustrated in Figure 14, avoids the risk of disturbing the application while the measurements are being run in lower levels. However, this optional optimization of resources MUST be used carefully.¶
The chosen period to measure a lower "qos-level" is implementation dependent. Therefore, it is not included as a "measurement:procedure" parameter. It is RECOMMENDED to use a large value, such as 20 minutes.¶
In order to allow peer-to-peer applications, a Q4S User Agent (UA) MUST be able to assume both the client and server role. The role assumed depends on who sends the first message.¶
In a communication between two UAs, the UA that first sends the Q4S BEGIN request to start the Handshake phase shall assume the client role.¶
If both UAs send the BEGIN request at the same time, they will wait for a random time to restart again as shown in Figure 15.¶
Otherwise, an UA may be configured to act only as server (e.g., content provider's side).¶
A Q4S session is intended to be used for an application. This means that for using the application, the client MUST establish only one Q4S session against the server. Indeed, the relation between the Session-Id and the application is 1 to 1.¶
If a user wants to participate in several independent Q4S sessions simultaneously against different servers (or against the same server), it can execute different Q4S clients to establish separately different Q4S sessions, but it is NOT RECOMMENDED because:¶
A Q4S client has different behaviors. We will use letters X, Y, and Z to designate each different behavior (follow the letters in Figure 16 and their descriptions below).¶
When it sends UDP messages (methods PING and BWIDTH), a Q4S client is not strictly a state machine that sends messages and waits for responses because of the following:¶
The Q4S-ALERT and CANCEL may have a conventional answer if an error is produced, otherwise the corresponding answer is formatted as a request message.¶
A valid Q4S request formulated by a client MUST, at a minimum, contain the following header fields:¶
At any time, if the server sends new SDP with updated values, the client MUST take it into account.¶
If a server does not understand a header field in a request (that is, the header field is not defined in this specification or in any supported extension), the server MUST ignore that header field and continue processing the message.¶
The role of the server is changed at Negotiation and Continuity phases, in which the server MUST send packets to measure jitter, latency, and bandwidth. Therefore, the different behaviors of the server are (follow the letters in Figure 17 and their descriptions below):¶
When the client begins to send UDP messages (methods PING and BWIDTH), a Q4S server is not strictly a state machine that receives messages and sends responses because of the following:¶
The Q4S-ALERT and CANCEL may have a conventional answer if an error is produced, otherwise the corresponding answer is formatted as a request message.¶
To provide a default configuration, it would be good if the client had a configurable set of quality headers in the implementation settings menu. Otherwise, these quality headers will not be present in the first message.¶
Different business models (out of scope of this proposal) may be achieved: depending on who pays for the quality session, the server can accept certain client parameters sent in the first message, or force billing parameters on the server side.¶
Different client and server implementations may send a different number of PING messages for measuring, although at least 255 messages should be considered to perform the latency measurement. The Stage 0 measurements may be considered ended only when neither the client nor server receive new PING messages after an implementation-dependent guard time. Only after, the client can send a "READY 1" message.¶
In execution systems, where the timers are not accurate, a recommended approach consists of including the optional Timestamp header field in the PING request with the time in which the message has been sent. This allows an accurate measurement of the jitter even with no identical intervals of time between PINGs.¶
In programming languages or operating systems with limited timers or clock resolution, it is recommended to use an approach based on several intervals to send messages of 1KB (= 8000 bits) in order to reach the required bandwidth consumption, using a rate as close as possible to a constant rate.¶
For example, if the resolution is 1 millisecond, and the bandwidth to reach is 11 Mbps, a good approach consists of sending:¶
1 message of 1KB every 1 millisecond + 1 message of 1KB every 3 milliseconds + 1 message of 1KB every 23 milliseconds¶
The number of intervals depends on the required bandwidth and accuracy that the programmer wants to achieve.¶
Considering messages of 1KB (= 8000 bits), a general approach to determine these intervals is the following:¶
To achieve the 375 messages per second, use a submultiple of 1000, which must be less than 375:¶
1000 / 2 = 500 > 375 1000 / 3 = 333 < 375¶
In this case, a message every 3 ms is suitable. The new pending target bandwidth is 375 - 333 = 42 messages per second.¶
Repeat the same strategy as point 3 to reach the pending bandwidth. In this case, 23 ms is suitable because of the following:¶
1000 / 22 = 45 > 42 1000 / 23 = 43 > 42 1000 / 24 = 41.6 < 42¶
We can choose 24 ms, but then we need to cover an additional 0.4 messages per second (42 - 41.6 = 0.4), and 43 is a number higher than 42 but very close to it.¶
In execution systems where the timers are not accurate, a recommended approach consists of checking at each interval the number of packets that should have been sent at this timestamp since origin and send the needed number of packets in order to reach the required bandwidth.¶
The shorter the packets used, the more constant the rate of bandwidth measurement. However, this may stress the execution system in charge of receiving and processing packets. As a consequence, some packets may be lost because of stack overflows. To deal with this potential issue, a larger packet is RECOMMENDED (2KB or more), taking into account the overhead produced by the chunks' headers.¶
Depending on the application nature and network conditions, a packet loss resolution less than 1% may be needed. In such cases, there is no limit to the number of samples used for this calculation. A trade-off between time and resolution should be reached in each case. For example, in order to have a resolution of 1/10000, the last 10000 samples should be considered in the packet loss measured value.¶
The problem of this approach is the reliability of old samples. If the interval used between PING messages is 50 ms, then to have a resolution of 1/1000, it takes 50 seconds, and a resolution of 1/10000 takes 500 seconds (more than 8 minutes). The reliability of a packet loss calculation based on a sliding window of 8 minutes depends on how fast network conditions evolve.¶
Q4S can be used as a mechanism to measure and trigger network tuning and application-level actions (i.e. lowering video bit-rate, reducing multiplayer interaction speed, etc.) in real time in order to reach the application constraints, addressing measured possible network degradation.¶
There are two scenarios in which Q4S can be affected by network problems: loss of Q4S packets and outlier samples.¶
Lost UDP packets (PING or BWIDTH messages) don't cause any problems for the Q4S state machine, but if TCP packets are delivered too late (which we will consider as "lost"), some undesirable consequences could arise.¶
Q4S does have protection mechanisms to overcome these situations. Examples:¶
Outlier samples are those jitter or latency values far from the general/average values of most samples.¶
Hence, the Q4S default measurement method uses the statistical median formula for latency calculation, and the outlier samples are neutralized. This is a very common filter for noise or errors on signal and image processing.¶
Q4S could be used in two scenarios:¶
One server:¶
It is the common scenario in which the client contacts the server to establish a Q4S session.¶
N servers:¶
In Content Delivery Networks and in general applications where delivery of contents can be achieved by different delivery nodes, two working mechanisms can be defined:¶
In order to solve the client-to-client scenario, a Q4S register function MUST be implemented. This allows clients to contact each other for sending the BEGIN message. In this scenario, the Register server would be used by peers to publish their Q4S-Resource-Server header and their public IP address to enable the assumption of the server role.¶
The register function is out of scope of this protocol version because different HTTP mechanisms can be used, and Q4S MUST NOT force any.¶
Because Q4S does not transport any application data, Q4S does not jeopardize the security of application data. However, other certain considerations may take place, like identity impersonation and measurements privacy and integrity.¶
Identity impersonation could potentially produce anomalous Q4S measurements. If this attack is based on spoofing of the server IP address, it can be avoided using the digital signature mechanism included in the SDP. The network can easily validate this digital signature using the public key of the server certificate.¶
Integrity of Q4S measurements under any malicious manipulation (such as a Man-in-the-Middle (MITM) attack) relies on the same mechanism, the SDP signature.¶
The Signature header field contains the signed hash value of the SDP body in order to protect all the SDP data, including the measurements. This signature not only protects the integrity of data but also authenticates the server.¶
This protocol could be supported over IPsec. Q4S relies on UDP and TCP, and IPsec supports both. If Q4S is used for application-based QoS, then IPsec is operationally valid; however, if Q4S is used to trigger network-based actions, then measurements could be incorrect unless the IPsec ports can be a target of potential action over the network (such as prioritizing IPsec flows to measure the new, upgraded state of certain application flows).¶
Any loss of connectivity may interrupt the availability of the Q4S service and may result in higher packet loss measurements, which is just the desired behavior in these situations.¶
In order to mitigate availability issues caused by malicious attacks (such as DoS and DDoS), a good practice is to enable the Q4S service only for authenticated users. Q4S can be launched after the user is authenticated by the application. At this moment, the user's IP address is known, and the Q4S service may be enabled for this IP address. Otherwise, the Q4S service should appear unreachable.¶
Q4S bandwidth measurement is limited to the application needs. It means that all available bandwidth is not measured, but only the fraction required by the application. This allows other applications to use the rest of available bandwidth normally.¶
However, a malicious Q4S client could restart Q4S sessions just after finishing the Negotiation phase. The consequence would be to waste bandwidth for nothing.¶
In order to mitigate this possible anomalous behavior, it is RECOMMENDED to configure the server to reject sessions from the same endpoint when this situation is detected.¶
If the ideas described in this document are pursued to become a protocol specification, then the code points described in this document will need to be assigned by IANA.¶
An assigned port would make possible a future Q4S-aware network capable of reacting by itself to Q4S alerts. A specific port would simplify the identification of the protocol by network elements in charge of making possible reactive decisions. Therefore, the need for a port assignment by IANA may be postponed until there is the need for a future Q4S-aware network.¶
Service Name: Q4S¶
Transport Protocol(s): TCP¶
Name: Jose Javier Garcia Aranda¶
Email: jose_javier.garcia_aranda@nokia.com¶
Name: Jose Javier Garcia Aranda¶
Email: jose_javier.garcia_aranda@nokia.com¶
This document has no IANA actions.¶
Many people have made comments and suggestions contributing to this document. In particular, we would like to thank:¶
Victor Villagra, Sonia Herranz, Clara Cubillo Pastor, Francisco Duran Pina, Michael Scharf, Jesus Soto Viso, and Federico Guillen.¶
Additionally, we want to thank the Spanish Centre for the Development of Industrial Technology (CDTI) as well as the Spanish Science and Tech Ministry, which funds this initiative through their innovation programs.¶