📄 rfc2815.txt
字号:
provide policing (function three), policing for IEEE networks is generally implemented at the edge of the network by a layer-3 device. When this policing is performed only at the edges of the network it is of necessity approximate. This issue is discussed further in [IS802FRAME].3.1. Context of admission control and delay bounds As described above, it is the combination of priority-based scheduling and admission control that creates quantified delay bounds. Thus, any attempt to quantify the delay bounds expected by a given traffic class has to made in the context of the admission control elements. Section 6 of the framework [IS802FRAME] provides for two different models of admission control - centralized or distributed Bandwidth Allocators.Seaman, et al. Standards Track [Page 6]RFC 2815 Int-Serv Mappings on IEEE 802 Networks May 2000 It is important to note that in this approach it is the admission control algorithm that determines which of the Int-Serv services is being offered. Given a set of priority classes with delay targets, a relatively simple admission control algorithm can place flows into classes so that the bandwidth and delay behavior experienced by each flow corresponds to the requirements of the Controlled-Load service, but cannot offer the higher assurance of the Guaranteed service. To offer the Guaranteed service, the admission control algorithm must be much more stringent in its allocation of resources, and must also compute the C and D error terms required of this service. A delay bound can only be realized at the admission control element itself so any delay numbers attached to a traffic class represent the delay that a single element can allow for. That element may represent a whole L2 domain or just a single L2 segment. With either admission control model, the delay bound has no scope outside of a L2 domain. The only requirement is that it be understood by all Bandwidth Allocators in the L2 domain and, for example, be exported as C and D terms to L3 devices implementing the Guaranteed Service. Thus, the end-to-end delay experienced by a flow can only be characterized by summing along the path using the usual RSVP mechanisms.3.2. Default service mappings Table 1 presents the default mapping from delay targets to IEEE 802.1 user_priority classes. However, these mappings must be viewed as defaults, and must be changeable. In order to simplify the task of changing mappings, this mapping table is held by *switches* (and routers if desired) but generally not by end-station hosts. It is a read-write table. The values proposed below are defaults and can be overridden by management control so long as all switches agree to some extent (the required level of agreement requires further analysis). In future networks this mapping table might be adjusted dynamically and without human intervention. It is possible that some form of network-wide lookup service could be implemented that serviced requests from clients e.g., traffic_class = getQoSbyName("H.323 video") and notified switches of what traffic categories they were likely to encounter and how to allocate those requests into traffic classes. Alternatively, the network's admission control mechanisms might directly adjust the mapping table to maximize the utilization of network resources. Such mechanisms are for further study.Seaman, et al. Standards Track [Page 7]RFC 2815 Int-Serv Mappings on IEEE 802 Networks May 2000 The delay bounds numbers proposed in Table 1 are for per-Bandwidth Allocator element delay targets and are derived from a subjective analysis of the needs of typical delay-sensitive applications e.g., voice, video. See Annex H of [802.1D] for further discussion of the selection of these values. Although these values appear to address the needs of current video and voice technology, it should be noted that there is no requirement to adhere to these values and no dependence of IEEE 802.1 on these values. user_priority Service 0 Default, assumed to be Best Effort 1 reserved, "less than" Best Effort 2 reserved 3 reserved 4 Delay Sensitive, no bound 5 Delay Sensitive, 100ms bound 6 Delay Sensitive, 10ms bound 7 Network Control Table 1 - Example user_priority to service mappings Note: These mappings are believed to be useful defaults but further implementation and usage experience is required. The mappings may be refined in future editions of this document. With this example set of mappings, delay-sensitive, admission controlled traffic flows are mapped to user_priority values in ascending order of their delay bound requirement. Note that the bounds are targets only - see [IS802FRAME] for a discussion of the effects of other non-conformant flows on delay bounds of other flows. Only by applying admission control to higher-priority classes can any promises be made to lower-priority classes. This set of mappings also leaves several classes as reserved for future definition. Note: this mapping does not dictate what mechanisms or algorithms a network element (e.g., an Ethernet switch) must perform to implement these mappings: this is an implementation choice and does not matter so long as the requirements for the particular service model are met. Note: these mappings apply primarily to networks constructed from devices that implement the priority-scheduling behavior defined as the default in 802.1D. Some devices may implement more complex scheduling behaviors not based only on priority. In that circumstance these mappings might still be used, but other, moreSeaman, et al. Standards Track [Page 8]RFC 2815 Int-Serv Mappings on IEEE 802 Networks May 2000 specialized mappings may be more appropriate.3.3. Discussion The recommendation of classes 4, 5 and 6 for Delay Sensitive, Admission Controlled flows is somewhat arbitrary; any classes with priorities greater than that assigned to Best Effort can be used. Those proposed here have the advantage that, for transit through 802.1D switches with only two-level strict priority queuing, all delay-sensitive traffic gets "high priority" treatment (the 802.1D default split is 0-3 and 4-7 for a device with 2 queues). The choice of the delay bound targets is tuned to an average expected application mix, and might be retuned by a network manager facing a widely different mix of user needs. The choice is potentially very significant: wise choice can lead to a much more efficient allocation of resources as well as greater (though still not very good) isolation between flows. Placing Network Control traffic at class 7 is necessary to protect important traffic such as route updates and network management. Unfortunately, placing this traffic higher in the user_priority ordering causes it to have a direct effect on the ability of devices to provide assurances to QoS controlled application traffic. Therefore, an estimate of the amount of Network Control traffic must be made by any device that is performing admission control (e.g., SBMs). This would be in terms of the parameters that are normally taken into account by the admission control algorithm. This estimate should be used in the admission control decisions for the lower classes (the estimate is likely to be a configuration parameter of SBMs). A traffic class such as class 1 for "less than best effort" might be useful for devices that wish to dynamically "penalty tag" all of the data of flows that are presently exceeding their allocation or Tspec. This provides a way to isolate flows that are exceeding their service limits from flows that are not, to avoid reducing the QoS delivered to flows that are within their contract. Data from such tagged flows might also be preferentially discarded by an overloaded downstream device. A somewhat simpler approach would be to tag only the portion of a flow's packets that actually exceed the Tspec at any given instant as low priority. However, it is often considered to be a bad idea to treat flows in this way as it will likely cause significant re- ordering of the flow's packets, which is not desirable. Note that the default 802.1D treatment of user_priorities 1 and 2 is "less than" the default class 0.Seaman, et al. Standards Track [Page 9]RFC 2815 Int-Serv Mappings on IEEE 802 Networks May 20004. Computation of integrated services characterization parameters by IEEE 802 devices The integrated service model requires that each network element that supports integrated services compute and make available certain "characterization parameters" describing the element's behavior. These parameters may be either generally applicable or specific to a particular QoS control service. These parameters may be computed by calculation, measurement, or estimation. When a network element cannot compute its own parameters (for example, a simple link), we assume that the device sending onto or receiving data from the link will compute the link's parameters as well as it's own. The accuracy of calculation of these parameters may not be very critical; in some cases loose estimates are all that is required to provide a useful service. This is important in the IEEE 802 case, where it will be virtually impossible to compute parameters accurately for certain topologies and switch technologies. Indeed, it is an assumption of the use of this model by relatively simple switches (see [IS802FRAME] for a discussion of the different types of switch functionality that might be expected) that they merely provide values to describe the device and admit flows conservatively. The discussion below presents a general outline for the computation of these parameters, and points out some cases where the parameters must be computed accurately. Further specification of how to export these parameters is for further study.4.1. General characterization parameters There are some general parameters [GENCHAR] that a device will need to use and/or supply for all service types: * Ingress link * Egress links and their MTUs, framing overheads and minimum packet sizes (see media-specific information presented above). * Available path bandwidth: updated hop-by-hop by any device along the path of the flow. * Minimum latency Of these parameters, the MTU and minimum packet size information must be reported accurately. Also, the "break bits" must be set correctly, both the overall bit that indicates the existence of QoS control support and the individual bits that specify support for a particular scheduling service. The available bandwidth should be reported as accurately as possible, but very loose estimates are acceptable. The minimum latency parameter should be determined and reported asSeaman, et al. Standards Track [Page 10]RFC 2815 Int-Serv Mappings on IEEE 802 Networks May 2000 accurately as possible if the element offers Guaranteed service, but may be loosely estimated or reported as zero if the element offers only Controlled-Load service.4.2. Parameters to implement Guaranteed Service A network element supporting the Guaranteed Service [GS] must be able to determine the following parameters: * Constant delay bound through this device (in addition to any value provided by "minimum latency" above) and up to the receiver at the next network element for the packets of this flow if it were to be admitted. This includes any access latency bound to the outgoing link as well as propagation delay across that link. This value is advertised as the 'C' parameter of the Guaranteed Service. * Rate-proportional delay bound through this device and up to the receiver at the next network element for the packets of this flow if it were to be admitted. This value is advertised as the 'D' parameter of the Guaranteed Service. * Receive resources that would need to be associated with this flow (e.g., buffering, bandwidth) if it were to be admitted and not suffer packet loss if it kept within its supplied Tspec/Rspec. These values are used by the admission control algorithm to decide whether a new flow can be accepted by the device. * Transmit resources that would need to be associated with this flow (e.g., buffering, bandwidth, constant- and rate-proportional delay bounds) if it were to be admitted. These values are used by the admission control algorithm to decide whether a new flow can be accepted by the device. The exported characterization parameters for this service should be reported as accurately as possible. If estimations or approximations are used, they should err in whatever direction causes the user to receive better performance than requested. For example, the C and D error terms should overestimate delay, rather than underestimate it.4.3. Parameters to implement Controlled Load A network element implementing the Controlled Load service [CL] must be able to determine the following: * Receive resources that would need to be associated with this flow (e.g., buffering) if it were to be admitted. These values are used by the admission control algorithm to decide whether a new flow can be accepted by the device.Seaman, et al. Standards Track [Page 11]RFC 2815 Int-Serv Mappings on IEEE 802 Networks May 2000 * Transmit resources that would need to be associated with this flow (e.g., buffering) if it were to be admitted. These values are used by the admission control algorithm to decide whether a new flow can be accepted by the device. The Controlled Load service does not export any service-specific characterization parameters. Internal resource allocation estimates should ensure that the service quality remains high when considering the statistical aggregation of Controlled Load flows into 802 traffic classes.4.4. Parameters to implement Best Effort For a network element that implements only best effort service there are no explicit parameters that need to be characterized. Note that an integrated services aware network element that implements only
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -