Lecture 14: Performance Architecture Prof. Shervin Shirmohammadi SITE, University of Ottawa Prof. Shervin Shirmohammadi CEG 4185 14-1 Background Performance: levels for capacity, delay, and RMA. Performance architecture is the set of mechanisms to configure, operate, manage, provision, and account for resources to support the required performance. Leads to Service Level Agreements (SLA), resource control, and Quality of Service (QoS) parameters. Typically higher performance means: higher network cost. more complicated routers more complicated applications more complicated protocols more suitable-trained human operators Example: For telelearning to remote areas, uottawa asks its ISP for guaranteed 300 Kbps video conferencing channel with maximum values of 150 msec and 5% for end to end delay and loss rate, respectively. Does this exist on the Internet? Prof. Shervin Shirmohammadi CEG 4185 14-2 1
SLA Example Service Levels: Capacity Performance Delay Performance Reliability Performance Basic Service As Available (Best Effort) As Available (Best Effort) As Available (Best Effort) Silver Service 1.5 Mb/s (Bidirectional) As Available (Best Effort) As Available (Best Effort) Gold Service 10 Mb/s (Bidirectional) (Burst to 100 Mb/s) Max 100-ms Roundtrip (Between Points) As Available (Best Effort) Platinum Service 100/10 Mb/s Up/down (Burst to 1 Gb/s) Max 40-ms Roundtrip (Between Points) 99.999% Uptime (User-Server) Prof. Shervin Shirmohammadi CEG 4185 14-3 Quality of Service: QoS Application s Quality of Service, as perceived by the user. Good quality: Smooth video, intelligible audio, natural conversation, easy interaction, Bad quality: Frozen video, unintelligible audio, difficult to interact and converse, For a specific applications, the above subjective perceptions are translated into objective parameters such as: network loss ratio Bandwidth variations delay jitter video size colour resolution Audio sampling rate Prof. Shervin Shirmohammadi CEG 4185 14-4 2
QoS Parameters Many studies have been performed to determine the thresholds for these parameters. E.g.: small video at 5 frames per second is sufficient for a talking head (video conference, news announcer, ) Frame rate of >25 fps is needed for movie-quality video End-to-end delay in a full-duplex audio conversation should be less than 150 msec In a synchronized audio/video presentation, the skew should be less than 15 msec For remote collaborative object manipulation, the delay should be less than 200 msec How to ensure such thresholds are met? Specially on the Internet! Network-level QoS (not implemented in the general Internet) Application-level QoS Prof. Shervin Shirmohammadi CEG 4185 14-5 Elastic Traffic Can adjust to changes in delay and throughput e.g. common TCP and UDP applications: email insensitive to delay changes FTP User expect delay proportional to file size Sensitive to changes in throughput SNMP delay not a problem, except when caused by congestion Web (HTTP) TELNET sensitive to delay, still considered elastic Total elapsed time: E.g. web page loading time For small items, delay across internet dominates For large items it is throughput over connection Prof. Shervin Shirmohammadi CEG 4185 14-6 3
Inelastic Traffic Does not easily adapt to changes in delay and throughput Real time traffic Throughput Minimum bandwidth may be required Delay E.g. stock trading, voice over IP Jitter: Delay variation More jitter requires a bigger buffer E.g. teleconferencing requires reasonable delay upper bound Packet loss Prof. Shervin Shirmohammadi CEG 4185 14-7 Issues with Inelastic Traffic Difficult to meet requirements on a network with variable queuing delays and congestion Need preferential treatment given by the network Can t do this in the transport layer Applications need to state requirements Preferably ahead of time, or on-the-fly Need some sort of resource reservation protocol Must still support elastic traffic Don t choke too much elastic traffic in favour of inelastic traffic People still have to check their email and surf the web Deny service requests that leave too few resources to handle elastic traffic demands Prof. Shervin Shirmohammadi CEG 4185 14-8 4
Integrated Services Architecture (ISA) IETF standard QoS support at the networking layer. Provision of QoS over IP Philosophy: Limit demand & reserve resources Sharing available capacity when congested non-isa Router mechanisms Routing Algorithms Select to minimize delay Packet discard Causes TCP sender to back off and reduce load This is enhanced by ISA Prof. Shervin Shirmohammadi CEG 4185 14-9 ISA Functions Admission control For QoS, reservation required for new flow uses RSVP Routing algorithm Routing decision based on QoS parameters Queuing discipline Take account of different flow requirements Fair queuing, processor sharing, weighted queuing, Discard policy Manage congestion Meet QoS Keep in mind, all of this is being done in the router! Prof. Shervin Shirmohammadi CEG 4185 14-10 5
ISA Implementation in Router Background Functions Forwarding functions Prof. Shervin Shirmohammadi CEG 4185 14-11 ISA Components Reservation Protocol RSVP Admission Control Enough resources to handle desired QoS? Management Agent As seen in NM lecture Routing Protocol Classifier and Route Selection Incoming packets mapped to classes Based on IP header fields Determines next hop Packet Scheduler Manages one or more queues for each output Order queued packets sent Policing Used for resource allocation Used for routing IP ToS field (1 byte) Prof. Shervin Shirmohammadi CEG 4185 14-12 6
ISA Services Traffic Specification (TSpec) defined as service for flow Becomes a sort of SLA On two levels General categories of service Guaranteed Controlled load Best effort (default) Particular flow within category TSpec is part of the contract between user and service provider. How to police an average data rate? One way: token bucket Prof. Shervin Shirmohammadi CEG 4185 14-13 Token Bucket Allows traffic to be sent in bursts, as long as the average data rate is not violated. Prof. Shervin Shirmohammadi CEG 4185 14-14 7
RSVP: Resource ReSerVation Protocol IETF standard Reserves resources on the routers (so, again, at the networking layer) of a given path, such that QoS requirements can be met. Enable receivers to make reservations on network routers Multicasting supported Simplex Unidirectional data flow Receivers can select one of multiple sources (channel) Deal gracefully with changes in routes Re-establish reservations Independent of routing protocol Uses the Type of Service field in the IPv4 header. Prof. Shervin Shirmohammadi CEG 4185 14-15 RSVP Operation Diagram Prof. Shervin Shirmohammadi CEG 4185 14-16 8
Differentiated Services (DS) Another IETF standard ISA and RSVP complex to deploy, and may not scale well for large volumes of traffic Amount of control signals Maintenance of state information at routers DS architecture designed to provide simple, easy to implement, low overhead tool Classify traffic in groups, and each group is handled differently Support a range of network services Differentiated on basis of performance Use IPv4 header Type of Service or IPv6 Traffic Class field No change to IP Service level agreement (SLA) established between provider (internet domain) and customer prior to use of DS DS mechanisms not needed in applications! (already negotiated) Build in aggregation: all traffic with same DS field treated same e.g. multiple voice connections DS implemented in individual routers by queuing and forwarding based on DS field State information on flows not saved by routers Prof. Shervin Shirmohammadi CEG 4185 14-17 DiffServ Traffic Classes Best Effort Typical to the best effort QoS we discussed before. Assured Forwarding (AF, RFC 2597) Used for traffic where both delay and capacity requirements need to be considered (tele-services). Four AF classes are defined, each with its own resources. Within each class, packets are marked with one of three drop precedence values. Those with lower drop precedence have higher priority. Expedited Forwarding (EF, RFC 2598) Targeted towards traffic that has strict delay requirements (real-time or interactive). Low loss, low-delay, and low-jitter end to end services through DS domains. Prof. Shervin Shirmohammadi CEG 4185 14-18 9
DS Field Uses the first 6 bits of IP ToS (the remaining 2 bits are unused). For AF Only PHB: Per hop behaviour Prof. Shervin Shirmohammadi CEG 4185 14-19 DiffServ vs. IntServ Function/Feature DiffServ IntServ Scalability Scalable to Large Enterprise of Service- Provider Networks Limited to Small or Medium Size Enterprise Networks Granularity of Control Traffic Aggregated into Classes Per-Flow or Groups of Flows Scope of Control Per Network Device (Per- Hop) All Network devices in End-to-End Path of Flow Prof. Shervin Shirmohammadi CEG 4185 14-20 10
Proactive Packet Discard A technique used in routers to avoid congestion before it happens. Discard packets before buffer is full Why would that reduce traffic? Used on single FIFO queue or multiple queues for elastic traffic e.g. Random Early Detection (RED) Prof. Shervin Shirmohammadi CEG 4185 14-21 Random Early Detection Traffic surges fill buffers and cause discards On TCP this is a signal to enter slow start phase, reducing load Lost packets need to be resent Adds to load and delay Global synchronization Traffic burst fills queues so packets are lost Many TCP connections enter slow start Traffic drops so network underutilized Connections leave slow start at same time causing burst Bigger buffers do not help (longer delays) Try to anticipate onset of congestion and tell one connection to slow down Prof. Shervin Shirmohammadi CEG 4185 14-22 11
RED Algorithm Calculate average queue size avg if avg < TH min queue packet else if TH min avg < Th max calculate probability P a with probability P a discard packet else with probability 1-P a queue packet else if avg TH max discard packet RED Buffer Prof. Shervin Shirmohammadi CEG 4185 14-23 12