1 / 25

INCITE I nter N et C ontrol and I nference T ools at the E dge

INCITE I nter N et C ontrol and I nference T ools at the E dge. R. Baraniuk R. Nowak E. Knightly R. Riedi M. Coates X. Wang V. Ribeiro S. Sarvotham. NMS PI meeting Atlanta October 2001. Chirp Probing. Effort 1. Objective : Reduced complexity, multiscale

zenda
Download Presentation

INCITE I nter N et C ontrol and I nference T ools at the E dge

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. INCITE InterNet Control and Inference Tools at the Edge R. Baraniuk R. Nowak E. Knightly R. Riedi M. Coates X. Wang V. Ribeiro S. Sarvotham NMS PI meeting Atlanta October 2001 Rice University, INCITE project incite.rice.edu

  2. Chirp Probing Effort 1 Rice University, INCITE project incite.rice.edu

  3. Objective: Reduced complexity, multiscale link models with known accuracy Innovative Ideas Multifractal analysis Multiplicative modeling Multiscale queuing Chirps for probing Impact Congestion control Workload balancing at servers Dynamical streaming Pricing on connection basis Rice University, INCITE project incite.rice.edu

  4. Chirp Probe Cross-Traffic Inference Rice University, INCITE project incite.rice.edu

  5. New Ideas Probing multiple hops Network calculus Probe size distributions Probing buffer at core router Passive inference (Sprint) Tech Transfer CAIDA(chirping as a monitoring tool) Stanford (SLAC) (chirps and PingER) Los Alamos (LANL) Sprint Labs Microsoft(streaming applications) UCRiverside(expertise on self-similarity) Rice University, INCITE project incite.rice.edu

  6. Connection-level Analysis and Modeling of Network Traffic Effort 2 Rice University, INCITE project incite.rice.edu

  7. 99% Mean Aggregate Statistics Auckland 2000 • Aggregate load on link • Time stamped headers • Positive process • Burstiness • LRD (large scale) • Non-Gaussian (small scale) Objective : Origins of small scale bursts Rice University, INCITE project incite.rice.edu

  8. Bursts in the ON/OFF framework • ON/OFF model • Superposition of sources • Connection level model • Explains large scale variability: • LRD, Gaussian • Cause: Costumers • Heavy tailed file sizes !! • Small scale bursts: • Non-Gaussianity • Conspiracy of sources ?? • Flash crowds ?? • (dramatic increase of active sources) Rice University, INCITE project incite.rice.edu

  9. 99% 99% Mean Mean Non-Gaussianity: A Conspiracy? Load: Bytes per 500 ms • The number of active connections is close to Gaussian; provides no indication of bursts in the load Number of active connections • Indication for: • - No conspiracy of sources • - No flash crowds Rice University, INCITE project incite.rice.edu

  10. Histogram of load offered in same time bin per connection: Considerable balanced “field” of connections Histogram of load offered in same time bin per connection: One connection dominates 10 Kb 150 Kb Non-Gaussianity: a case study Typical Gaussian arrival (500 ms time slot) Typical bursty arrival (500 ms time slot) Rice University, INCITE project incite.rice.edu

  11. 99% Mean Non-Gaussianity and Dominance • Dominant connections correlate with bursts Circled in Red: Instances where one connections contributes over 50% of load(resolution 500 ms) Rice University, INCITE project incite.rice.edu

  12. 99% Mean Non-Gaussianity and Dominance Systematic study: time series separation • For each bin of 500 ms: remove packets of the ONEstrongest connection • Leaves “Gaussian” residual traffic = + Overall traffic 1 Strongest connection Residual traffic Rice University, INCITE project incite.rice.edu

  13. Separation on Connection Level Definition: • Alpha connections: Peak rate > mean arrival rate + 1 std dev • Beta connections: Residual traffic • Findings are similar for • Auckland (2000+2001), Berkeley, Bellcore, DEC • 500ms, 50ms, 5ms resolution Rice University, INCITE project incite.rice.edu

  14. Alpha Traffic Component • There are fewAlpha connections • < 1% (AUCK 2000: 427 of 64,087 connections) • 3% of load Multifractal spectrum: Wide spectrum means bursty • Alpha connections • cause bursts: Alpha is extremely bursty Beta is little bursty Overall traffic is quite bursty Balanced (50% alpha) very bursty Rice University, INCITE project incite.rice.edu

  15. 99% 99% Mean Mean Beta Traffic Component • Constitutes main load • Governs LRD properties of overall traffic • Is Gaussian at sufficient utilization (Kurtosis = 3) • Is well matched by ON/OFF model Variance time plot Beta traffic Number of connections = ON/OFF Rice University, INCITE project incite.rice.edu

  16. What Causes Alpha Connections? • Potential causes: • TCP slow-start peculiarities • Start/End of “massive” flows • Re-routing • Heterogeneity in bandwidths • Look for: systematic explanation • First two: anywhere in network • Last two: locality in network important Rice University, INCITE project incite.rice.edu

  17. Origins of Alpha Traffic 1 • Observation 1: Alpha connections cluster into e2e groups • e2e group: connections with same source-receiver pair • 85 (out of total 6960) e2e groups contain at least one of the 427 alpha connections (AUCK) • Locality matters  Excludes TCP slow start and start/end of “massive” flows as systematic causes Rice University, INCITE project incite.rice.edu

  18. Origin of Alpha Traffic 2 • Observation 2: If one connection in e2e-group is alpha, then all connections are • Unlimited (Peak rate > ½ Total transfer) • TCP control mechanism does not become effective and/or • Alpha (Peak > Threshold) • Causes burst Peak rate and total load per connection for two e2e groups Alpha threshold Alpha threshold Rice University, INCITE project incite.rice.edu

  19. Alpha threshold = 30 Alpha threshold Alpha threshold Origin of Beta traffic • Observation 3: If no connection in e2e-group is alpha, then all connections are • Limited by same bottleneck bandwidth Evidence for: bandwidth matters Confirms ON/OFF as a good model of Beta traffic Rice University, INCITE project incite.rice.edu

  20. Bursts arise from large transfers over fast links. Simple Connection Taxonomy This is the only systematic reason Rice University, INCITE project incite.rice.edu

  21. Modeling NetworkTraffic Physical Model • Traffic (user): superposition of ON/OFF sources requesting files with heavy tailed size • Network: heterogeneous bandwidth  variable sending-rates(fixed per ON/OFF source) • Explains properties of traffic: • LRD: heavy tailed transfer of beta sources (crowd) • Bursts: few large transfers of few alpha sources • Mathematical Model • Traffic = Alpha + fractional Gaussian noise Rice University, INCITE project incite.rice.edu

  22. Impact 1: Simulation • ns: topology should include a few alpha links Realistic: heterogeneous end-to-end bandwidth Simple: equal bandwidth Rice University, INCITE project incite.rice.edu

  23. Impact 2: Queueing • Beta: rules small queues • Alpha: rules the rare extremely long queues • Needs theoretical work Rice University, INCITE project incite.rice.edu

  24. Summary • Connection level analysis of “all” available traces • Typically onedominant connection during burst • Alpha traffic (peak rate > burst threshold) • Few connections. Responsible for bursts • Origin: Large transfer over highbandwidth paths • Bursts are less pronounced at high utilization • Beta traffic (residual): • Main load. Responsible for LRD • Origin: Crowd with limited bandwidth • Gaussian at sufficient utilization Rice University, INCITE project incite.rice.edu

  25. Future work • Queueing analysis • Mathematical model for Alpha • Further verification using • More (new) traces • Simulation (ns, testbed) Realism • Monitoring of real network • Influence of other parameters on presence of bursts • Utilization • Delay • Topology incite.rice.edu Rice University, INCITE project incite.rice.edu

More Related