170 likes | 255 Views
Measurement based traffic engineering. Poul Heegaard, Telenor R&D / NTNU Dept. Telematics. Resource allocation. Performance optimisation. Measurements. Test labs. Production networks. Open architecture. Performance guarantee. Service differentiation.
E N D
Measurement based traffic engineering Poul Heegaard, Telenor R&D / NTNU Dept. Telematics 1
Resource allocation Performance optimisation Measurements Test labs Production networks Open architecture Performance guarantee Service differentiation Gilb’s Law:“Anything can be measured in a way that is superior to not measuring it at all.” 2
Open architecture Performance guarantee Service differentiations Resource allocation Performance optimisation Measurements Test labs Production networks Foster rich set applications Worldwide commercial interests New actors protect investments New cost models QoS guarantees New QoS requirements New applications, New markets New QoS requirements 3
Open architecture Performance guarantee Service differentiations Measurements Test labs Production networks (core/AS, access/SPE/CPE) Controllable (Protect, Priority, Guarantee) Resource control Performance optimisation Proactive (planning, designing) Reactive (monitoring, reconfig) 4
Open architecture (service innovation, fairness) Performance guarantee Service differentiation Resource allocation Performance optimisation Test labs Production network • SLA fulfilled? • Mechanisms effective? • Effect of new applications? • New applications appeared? • Performance bottlenecks? • Connectivity? • Routing stability? • “Clever” users? • “Malicious” users? • Charging? • Measurements • domain (inter, intra, access, private) • level (physical, network, transport, application) • approach (active, passive) 5
Open architecture Performance guarantee Service differentiations Resource allocation Performance optimisation Measurements Test labs • redesign, configuration • connectivity • performance assurance • traffic trends • security • input to traffic modelling Production networks Essential: multipurpose measurement architecture 6
Multiple measurement objectives Planning / Long term operation IP Network QoS & Performance management MEASUREMENT FUNCTIONS Medium / Short term operation Multipurpose measurement probe Prognosis Trends Traffic matrix Overload (observation and control) DependabilityReliability Accounting Security Fraud SLA validation Measurement Application 7
End-to-end delay, delay variation (“jitter”), packet loss ratio Many options DoS attack? traffic trends? resource (e.g. link) utilisation? Asymmetric traffic? Volume per customer? DoS attack? DoS attack? traffic volume? ADSL 1 - POP Our own network Paradigm shift in networking: same platform for all services Should also apply to monitoring: same platform for all measurement needs Peering network less than 5% overhead DoS attack? Traffic volume (per customer)? service usage? Service provider 8
Open architecture Performance guarantee Service differentiations Resource allocation Performance optimisation Measurements Test labs • coarse grained data collection • imprecise active tests • performance demanding • excessive measurement data • increasing measurement needs • measurements by 3rd party • measurement architecture hard Production networks Essential: configurable, precise, up-to-date, available data 9
Ex: Delay estimation • two-way by single point (e.g. in tcp, rtp flows) • one-way by dual point (e.g. inexpensive probes) compare single and dual point estimations protocol effects no congestion (e.g. delayed ack) => perfect match congestion upper bounds on network layer delay 10
Ex: flow-sampling • Data reduction by sampling 11
Validation of delay measured by active tests 1000 900 800 700 600 [ms] 500 400 300 corrected packet delay 200 100 0 0,5 0,6 0,7 0,8 0,9 1 1,1 1,2 1,3 Load level for offered tcp traffic Ex: corrected active tests Measure quality when service is in use customer view Measure quality any time provider view 12
Open architecture Performance guarantee Service differentiations Resource allocation Performance optimisation Measurements Production networks Test labs • equipment • mechanisms (e.g. QoS) • configurations • applications • user behaviour Essential: realistic traffic generator => e.g. GenSyn 13
User behaviour model User behaviour model Internet protocols Internet protocols GenSyn - objectives • Controllable • Scalable • Re-producible • Realistic traffic New services New network mechanisms 14
GenSyn – in short • Java-based, portable traffic generator • Flexible and scalable • Stochastic state models of user behaviour • Link to protocol stack for real packet generation 15
Conflicting and interdependent interests Gordian Knot QoS guarantee differentiation security simplisity openess revenue protection real-time monitoring resource control segmentation 16