A Case for Flow-Adaptive Wireless Links
Reiner Ludwig
EECS Department, University of California, Berkeley
Technical Report No. UCB/CSD-99-1053
, 1999
http://www2.eecs.berkeley.edu/Pubs/TechRpts/1999/CSD-99-1053.pdf
We study the performance problems that exist when loss responsive flows traverse wireless links, where losses are often unrelated to congestion. We present a novel concept -- flow-adaptive wireless links -- which provides service differentiation by tailoring link layer error control to the QoS requirements of each flow sharing the link. Flow-adaptive links emphasize local error control as a necessary complement to end-to-end error control, and are independent of transport (or higher) layer protocol semantics. The key idea is that applications use the IP layer as a level of indirection through which QoS requirements are communicated to each link along the path, on a per flow basis. We then demonstrate how this improves performance for the particular class of reliable loss responsive flows. We prove in general that a well engineered, fully reliable wireless link does not interfere with TCP's end-to-end error recovery. Moreover, we propose a new error recovery algorithm (TCP-Eifel) that can optionally be implemented in TCP to further improve performance. By eliminating the retransmission ambiguity problem the algorithm detects spurious timeouts, and uses these as an implicit cross-layer signal to prevent unnecessary retransmissions in TCP.
BibTeX citation:
@techreport{Ludwig:CSD-99-1053, Author= {Ludwig, Reiner}, Title= {A Case for Flow-Adaptive Wireless Links}, Year= {1999}, Month= {May}, Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/1999/5518.html}, Number= {UCB/CSD-99-1053}, Abstract= {We study the performance problems that exist when loss responsive flows traverse wireless links, where losses are often unrelated to congestion. We present a novel concept -- flow-adaptive wireless links -- which provides service differentiation by tailoring link layer error control to the QoS requirements of each flow sharing the link. Flow-adaptive links emphasize local error control as a necessary complement to end-to-end error control, and are independent of transport (or higher) layer protocol semantics. The key idea is that applications use the IP layer as a level of indirection through which QoS requirements are communicated to each link along the path, on a per flow basis. We then demonstrate how this improves performance for the particular class of reliable loss responsive flows. We prove in general that a well engineered, fully reliable wireless link does not interfere with TCP's end-to-end error recovery. Moreover, we propose a new error recovery algorithm (TCP-Eifel) that can optionally be implemented in TCP to further improve performance. By eliminating the retransmission ambiguity problem the algorithm detects spurious timeouts, and uses these as an implicit cross-layer signal to prevent unnecessary retransmissions in TCP.}, }
EndNote citation:
%0 Report %A Ludwig, Reiner %T A Case for Flow-Adaptive Wireless Links %I EECS Department, University of California, Berkeley %D 1999 %@ UCB/CSD-99-1053 %U http://www2.eecs.berkeley.edu/Pubs/TechRpts/1999/5518.html %F Ludwig:CSD-99-1053