1 / 21

ProbeCast: MANET Admission Control via Probing

ProbeCast: MANET Admission Control via Probing. Soon Y. Oh, Gustavo Marfia, and Mario Gerla Dept. of Computer Science, UCLA Los Angeles, CA 90095, USA {soonoh, gmarfia, gerla}@cs.ucla.edu. Introduction. Multicast “inelastic” streams

Download Presentation

ProbeCast: MANET Admission Control via Probing

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. ProbeCast: MANET Admission Control via Probing Soon Y. Oh, Gustavo Marfia, and Mario Gerla Dept. of Computer Science, UCLA Los Angeles, CA 90095, USA {soonoh, gmarfia, gerla}@cs.ucla.edu

  2. Introduction • Multicast “inelastic” streams • Inelastic flow - the rate cannot be elastically controlled (unlike TCP) • Real time flows: situation awareness dissemination; surveillance data/video, etc • Important in tactical/emergency MANETs • Traditional resource reservation ineffective in MANETs • Bookkeeping is very cumbersome in multicast (as number of destinations increases); • Also, mobility requires continuous re-adjustments • Without reservations: • Flow allocation can be “unfair”  possible capture • Network may get congested

  3. Unfairness Example (unicast) • 3 parallel inelastic flows; 500Kbps each • Interference between Flow 1 and 2 and Flow 2 and 3

  4. Goal & Contribution • Achieving reliable QoS support of inelastic flows (e.g., video and audio stream) • ProbeCast: • Enable Call Admission Control and fair allocation of inelastic flows in MANETs without requiring prior resource reservation

  5. ProbeCast: key insights • Insight #1: Resource Probing • No a priori resource allocation • Rather “probe” for resources to see if available • Insight #2: Pruning via Back-pressure • Back-pressure (“prune”) toward the source when resource is unavailable • Re-route or reject the inelastic flow • Insight #3: Neighborhood Proportional Drop (NPROD) • Local rate balancing using proportional dropping • Enforces fair channel sharing  “fair back-pressure”

  6. ProbeCast: Example Backpressure (Pruning) Proportional Drop

  7. ProbeCast: Probing • Assumptions: • End-to-End FEC – e.g. erasure coding – always ON • Each flow has packet drop threshold (say, 20%), beyond which the flow must be back-pressured • Probing • Each node measures own packet drop rate • It broadcasts to one hop neighbors own drop rate via piggybacking on data packets

  8. The node estimates packet drop probability DPF for each flow F • It broadcasts to one hop neighbors the DPF value 8

  9. ProbeCast: N-PROD • Neighborhood Proportional Drop (N-PROD) • Distributed fairness scheme • First introduced and evaluated in FairCast (MSWIM 2008) • Overhearing neighbors’ drop probabilities • Enforcing proportional drop among flows competing in the same contention domain • Forced drop from the queue • After transient, nodes in the same contention domain converge to fair share of the channel

  10. ProbeCast: Pruning • Pruning • Flow Drop based on Threshold • Threshold is traffic class and flow age dependent; • Drop Threshold stamped in packet header • Typically, incoming flow has lower threshold than incumbent • When drop rate is > threshold, a flow is backpressured • BckPr signal piggybacked on data packets whenever possible • Upstream node in turn will backpressure when all “children” have sent BckPr signal • Source action (upon receiving backpressure signal): • Re-route if there is alternate route; • Otherwise reject the flow

  11. ProbeCast Example • Three flows in the same contention domain. Bar graphs shows packet delivery ratios • Flow 3 starts transmitting and other flows’ rate decreases (N-PROD). • Since Flow 3 drop rate exceeds the threshold, it is backpressured.

  12. Simulation • Simulation setup • Qualnet simulations • Radio range 376m; 2Mbps capacity; 802.11b • 512B packets; 50KB queue in each node • Topologies • Three parallel flows • 30 nodes uniformly distributed in a 1000x1000m field • Experiments • N-PROD: to show proportional fairness • ProbeCast: to show proper rejection

  13. S3 Source Flow 3 Forwarder F3 R3 Receiver Flow 2 S2 F2 R2 Flow 1 F1 R1 S1 Three Parallel Flow Topology • F1, F2, and F3 are within the same contention domain • No interference between sources and forwarders • No interference between forwarders and receivers • Staggered Transmission starts: 1s, 10s, 20s

  14. Three Even Parallel Flows Uniform nominal rate = 500Kpbs Flows 2 has higher packet drop rate without N-PROD N-PROD restores fairness 14

  15. Three Even Parallel Flows (cont) Uniform nominal rate = 500Kpbs Aggregated throughput of the three flows Fairness comes at the cost of degraded total throughput 15

  16. Three Uneven Parallel Flows Flow1 = 800Kbps, Flow2 = 400Kbps, and Flow3=200Kbps Without N-PROD, Flow 1 and 3 capture the channel With N-PROD proportional drop yields 8:4:2 ratio 16

  17. Two Flows in Random Topology Session 1 Session 2 • 30 nodes in 1000 by 1000 meter • Flow 1: 200Kbps video stream, 9 members • Flow 2: 40Kbps audio stream, 3 members Session 2

  18. Two Flows in Random Topology • Session 1 captures channel in ODMRP so session 2 starves • N-PROD achieves fairness • All members in session 1 and 2 receive more than 50% packets • Drop Threshold = 50%

  19. Three Flows in Random Topology • Three multicast sessions, each session has 1 source and 3 members • 30 nodes in 1000 by 1000 meter • Data transmission starts Session1 T=1s, Session2 T=10s, and Session3 T=20s • 500Kbps traffic; drop threshold = 50% Session 1 Session 2 Session 3

  20. Three Flows in Random Topology (cont) • Three multicast sessions compete within the same collision domain • Session 2 is rejected (it came after Session 1 - initially lower threshold)

  21. Conclusion • N-PROD achieves proportional bandwidth share in the same contention domain • ProbeCast uses probing and backpressure to accept feasible flows and reject unfeasible ones. • Probecast can also handle inelastic unicast (a special case of multicast)

More Related