End-to-End Performance and Fairness in Multihop Wireless Backhaul Networks V. Gambiroza, B. Sadeghi, and E. Knightly Department of Electrical and Computer Engineering Rice University MobiCom’04, Sept. 26Oct. Presented by Yeong-cheng Tzeng

Outline Introduction Performance and fairness objectives Performance study Capacity and fairness Conclusions TAPs Network

Introduction Wireless LANs remain slower due to slow wired backhaul Develop wireless backhaul networks via wirelessly multi- hopping to a Internet entry point Existing protocols result in –  Unfairness  Starvation  Poor performance

TAP System Model System wide performance Internet Backbone TAP Network Introduction (cont.)

TAP System Model System wide performance  If branches are on different frequencies or sufficiently spatially separated Internet Backbone TAP Network Introduction (cont.) Internet TAP1 TAP2 TAP3 TAP4

Introduction (cont.) Parking Lot Scenario  Similar to parking lot with one exit Internet TAP1 TAP2 TAP3 TAP4

Introduction (cont.) Parking Lot Scenario  Fairness problem

Introduction (cont.) Parking Lot Scenario  Fairness problem

Introduction (cont.) Parking Lot Scenario  Fairness problem

Introduction (cont.) Parking Lot Scenario  Fairness problem Goal Ensure equal shares independent of spatial location

Outline Introduction Performance and fairness objectives  Objectives  TAP fairness reference model Performance study Capacity and fairness Conclusions TAPs Network

Objectives Temporal fairness  vs. throughput fairness  Performance isolation Ingress aggregate  Each TAP corresponds to a single residence, small business, or hot spot Spatial bias  No penalty for flows further away from the wired TAP Spatial reuse  Resources can be reclaimed when they are unused  Maximize

TAP fairness reference model Present a formal definition that determines if a set of candidate allocated temporal shares (expressed as a matrix T) is TAP-fair  Contention neighborhood  A subset of the set of all links with the property that no two links from the subset can be active simultaneously  A matrix T satisfying following constraints is said to be feasible , for all flows (i, j) , for all k and all links n

TAP fairness reference model (cont.) Definition 1. A matrix T is said to be TAP fair if it is feasible and if for each flow, cannot be increased while maintaining feasibility without decreasing for some flow for which , when

TAP fairness reference model (cont.) Considering temporal shares we ensure temporal fairness We ensure the ingress aggregate objective by satisfying the below inequality , when We ensure the spatial bias objective by satisfying the below equality  By allowing no spare “ time capacity ” we ensure spatial reuse 

Outline Introduction Performance and fairness objectives Performance study  UDP Baseline Scenario  TCP fairness  Inter-TAP Fairness Algorithm (IFA)  Summary of Findings Capacity and fairness Conclusions TAPs Network

Performance Study Goal  Study end-to-end performance and fairness Factors investigated  Fairness algorithms Uncontrolled UDP, TCP, IFA  Media access control CSMA and CSMA/CA  Antenna technologies Omni directional, sector  Carrier sense range, multiple topologies and flow scenarios …

UDP Baseline Scenario Parking lot scenario MU-TAP and TAP-TAP transmissions on orthogonal channels 5 Mobile Users (MU) per TAP Constant rate UDP traffic with 1000 byte packets Channel rate constant 2 Mb/sec TAPs two or more hops away not in carrier sense range Internet TAP1 TAP2 TAP3 TAP4 TA(1) TA(2) TA(3)

UDP Baseline Scenario (cont.) Traffic originating at TAP1 starved  “Hidden terminal” problem An increased no. of hops leads to a corresponding throughput decrease UDP/ achieved 92% of the objective  Capacity and fairness need to be considered jointly No multi-rate transmission  Temporal = throughput

Fairness with TCP TAP4 ’ s ACK traffic Hidden terminal problem RTS/CTS introduces information asymmetry problem CSMA obtains slightly higher goodput than CSMA/CA TCP traffic is higher than objective Internet TAP1 TAP2 TAP3 TAP4 TA(1) TA(2) TA(3)

Fairness with TCP (cont.) Sector antennas eliminate the hidden terminal problem and information asymmetry problem Total goodput is increased  TAP(1) and TAP(2) are not starved  Second air interface  Reduced collisions Internet TAP1 TAP2 TAP3 TAP4 TA(1) TA(2) TA(3)

Fairness with TCP (cont.) A sensitive carrier sense range can mitigate the impact of hidden terminals and information asymmetry Such a sensitive carrier sense range not always true:  Realistic - due to hardware limitations  Desirable - reduces spatial reuse TAP1 TAP2 TAP3 TAP4

Inter-TAP Fairness Algorithm (IFA) A layer 2 multi-hop wireless fairness algorithm  Allocate resources according to reference model  By limiting flows at the first hop to mitigate unfairness and starvation Key components  Measurement of Offered Load and Capacity Measurement can be performed at TAPs or MUs  Message Distribution Message distribution interval Overhead: Control message require priority or spare capacity  Aggregate Fair Share Computation Compute the aggregate time shares in each contention neighborhoods and chooses the minimum value Convert the time share to rate via use of the available link capacity  Ingress Rage Limiting The TAP can signal each MU of its fair share The TAP can CTS or poll MUs at the desired rate

Inter-TAP Fairness Algorithm (IFA) (cont.)

End-to-end performance considerably improved  Hidden terminal problem mitigated  Contention considerably decreased Spatial bias  IFA cannot eliminate it Spatial reuse  IFA is able to exploit spatial reuse Rates lower than the objective

Summary of Findings “Parking Lot” scenario results in hidden terminals and information asymmetry  Starvation of upstream flows (UDP, TCP, with or w/o RTS/CTS) Sector antennas and carrier sense range mitigate the problem  Sector antennas: Throughput as low as 26% of targeted values  Sensitive CSR not always realistic and desirable Ingress rate limiting mitigates the problem

Outline Introduction Performance and fairness objectives Performance study Capacity and fairness  Maximum throughput without fairness  Fairness objectives and throughput  Throughput comparison Conclusions TAPs Network

Maximum throughput without fairness Assign time-shares to maximize network throughput No spare time-capacity

Maximum throughput without fairness (cont.) Solution:  Assign time shares to only one flow Preservation property

Fairness objectives and throughput Temporal Fairness Constraint Time is resource  Divided equally among flows

Example policy: Fairness objectives and throughput (cont.) Spatial Bias removal constraint Equal share regardless of number of hops for all *, ** Internet TAP1 TAP2 TAP3 TAP4 * : Fair share of flow f, defined by system policies **

Fairness objectives and throughput (cont.) Throughput is proportional to the throughput of ingress link

Fairness objectives and throughput (cont.) 20 km v1=20km/h 1hr v2=10km/hr 2hr AB

Ingress Aggregate flow granularity Ingress-Egress flow granularity Fairness objectives and throughput (cont.) Ingress Aggregate Constraint Treat same-ingress flows as one and

Fairness objectives and throughput (cont.)

Throughput comparison Example scenario Summary  Total throughput  Temporal vs. throughput  Throughput highly probable depending on link capacity 10Mbps flow(1,3) flow(1,2) TA(2) TA(3) 20Mbps5Mbps

Outline Introduction Performance and fairness objectives Performance study Capacity and fairness Conclusions TAPs Network

Conclusions Fairness  Fairness reference model formally defined  Designed for multi-hop wireless networks Performance study  Starvation of upstream flows  Sector antennas, larger carrier sense range, IFA mitigate the problem  IFA approximates performance of reference model Capacity and fairness  Need to be considered jointly

The End