? The instability of ad hoc routing badly affects TCP performance over 802.11 multihop networks ? We consider improving TCP performance by stabilizing routing dynamics over 802.11 networks
Electrical Engineering Department UNIVERSITY OF SOUTHERN CALIFORNIA
USC Sensitivity to Link Failure ( -factor) – Generalized sensitivity parameter to link failure– Response to link failure only for a number ( ) of successive link failures
Improving Stability and Performance of Multihop 802.11 Networks Kitae Nahm, Ahmed Helmy, C.-C. Jay Kuo helmy@usc.edu,{nahm,cckuo}@sipi.usc.edu
We introduce a simple reaction policy of DSR protocol to link failure in 802.11 networks We study the tradeoff of the proposed policy with respect to node mobility, channel error, and MAC contention loss
Node mobility – Node is no longer available– A new route should be established no matter how it might cost
– No retransmission of the lost packets by DSR protocol
The cross-layer interaction among 802.11 MAC, transport, and network layers are unorganized and inefficient
Wireless channel error – Channel error is a just matter of connection quality, not of the end-to-end connectivity– Without a better alternative path, channel error should be tolerated– Link-layer retransmission can help if needed
– TCP pushes as many packets as possible into 802.11 network, which intensifies contention in wireless medium– Extended hidden/exposed terminal problem[1,2]
*Not necessarily with the same packets
TCP drives on-demand ad hoc routing unstable in 802.11 networks – DSR protocol performs routing maintenance and rediscovery operations by taking link loss for node mobility and wireless channel error– TCP congestion (MAC contention loss) misleads routing agent into unnecessary routing maintenance operations F TCP connection failure
Routing Stability and TCP Throughput
G TCP timeout
C Routing failure
Maintenance traffic (Re-routing, error messaging, etc)
D
B
MAC loss (Congestion)
E Network overload
Routing changes in a static 7x7 grid topology TCP restart
The route was changed over 100 times during the simulation session, and the shortest path is utilized only under 50% The route was changed only a single time and the 6-hop shortest path was utilized over 85%
No channel noise (PER 0.0) DSR (=2) almost achieves the TCP performance bound of the static routing DSR (=1) achieves about 50% TCP throughput of the static routing
Manhattan mobility model – 200 mobile nodes in 500x500m space– Transmission range 60m, carrier sensing/interference range 140m
PER (packet error rate) 0.04 – DSR (=2) achieves over 200% improvement of TCP throughput of DSR (=1) in 420 hop chain topologies
Conclusions β
The on-demand ad hoc routing protocols can be extremely inefficient in 802.11 networks because different types of networking events such as wireless channel error, node mobility, and congestion are treated in the same way Our proposed reaction policy applied to DSR protocol noticeably improves the performance and stability of 802.11 networks without any effort to distinguish those events
Network Events Node Mo
bility Congestion Channel Error
Tolerability of Link Failure Intolerable Should be tolerated Tolerable
Impact of>1+ 0
[1] Z. Fu, et al.“The impact of multihop wireless channel on TCP throughput and loss”, INFOCOM 2003[2] K. Chen, et al.“Understanding bandwidth-delay product in mobile ad hoc networks”, Computer Communications, 2004[3] K. Nahm, et al.“TCP over multihop 802.11 networks: issues and performance enhancement”, ACM MobiHoc, May, 2005
NS-2 code will be available soon at doc.guandang.net
β
TCP vulnerability to channel error is the same, but the improvement of routing robustness with=2 is more beneficial to the overall system performance including TCP throughput
The impact of the increased mobility-reaction latency with=2 on the overall system performance is practically negligible
β
=2 makes noticeable difference in low mobility (< 5m/s) scenarios where TCP throughput is practically meaningful – 65-85% less route changes– 35-200% higher TCP throughput
β
-factor Tradeoff: Wireless Channel Error
β
–
Connection blackout in chain topologies for the case of DSR protocol
cycle[3]
DSR (=2) experienced has almost the same throughput as static routing
β
A
Network overload initiated by TCP
H
DSR (=1, original) has 13% lower throughput than static routing –
-factor Tradeoff: Node Mobility
Reference
β
TCP drives MAC contention loss instead of queuing loss in 802.11 networks
β
β
Lack of coordination in sharing network resource –
– Congestion is a part of TCP operation– The current route should remain intact
– The conventional DSR implementation
– DSR protocol responds after 7 successive retransmissions in 802.11 MAC*– Pro: suppressing unnecessary routing at congestion– Con: slower mobility response
β
Cross-layer Interaction in 802.11 Networks
Congestion
L: the number of successive link failures L<: The current packet is dropped and the transmission is resumed for the next packet L=: Routing maintenance/re-discovery. Reset L=0
=1>1
β
MAC loss needs to be treated differently depending on the types of the associated networking events– On-demand routing is unstable with TCP because wireless channel error, node mobility, and congestion are treated in the same way
β
The instability of ad hoc routing badly affects TCP performance over 802.11 multihop networks We consider improving TCP performance by stabilizing routing dynamics over 802.11 networks –
Introduction of
factor for DSR protocol
β
Introduction
Tolerability of 802.11 Link Failure
β
β
β
β
β
β
β
–
–
–

, ?, § ) Department of Electrical Engineering, ?Department of Chemistry, ... University of Southern California, Los Angeles, California 90089, United ...
University of Southern California(简写 USC), 建于 1880 年,位于加利福尼亚州的...Biomedical Engineering (Electrical Engineering) (B.S.) Biomedical Engineering ...
, ?, § ) Department of Electrical Engineering, ?Department of Chemistry, ... University of Southern California, Los Angeles, California 90089, United ...
热门文档