Protocols that provide reliable communication on top of a network that can lose packets rely on periodically retransmitting packets. The choice of retransmission timeout critically affects system performance. This paper presents a first step toward a theoretical study of the choice of retransmission timeout, based on competitive analysis. In general, competitive analysis compares the performance of an on-line algorithm to the performance of an optimal off-line algorithm, which has access to more information. In this context, the job of an algorithm is to choose the retransmission timeout interval; an off-line algorithm knows the exact message delays, whereas an on-line algorithm knows only upper and lower bounds on the delays. The performance measure of interest is the expected value of a linear combination of the number of packets used and the amount of time elapsed. An on-line algorithm for choosing the retransmission timeout is presented that is optimal with respect to the difference between its performance and that of an optimal off-line algorithm. The algorithm is also analyzed with respect to the ratio of its performance and that of an optimal off-line algorithm.
[1]
Yuval Rabani,et al.
Competitive Algorithms for Distributed Data Management
,
1995,
J. Comput. Syst. Sci..
[2]
Allan Borodin,et al.
An optimal online algorithm for metrical task systems
,
1987,
STOC.
[3]
Lyle A. McGeoch,et al.
Competitive algorithms for on-line problems
,
1988,
STOC '88.
[4]
Robert E. Tarjan,et al.
Amortized efficiency of list update and paging rules
,
1985,
CACM.
[5]
Raj Jain.
Divergence of Timeout Algorithms for Packet Retransmissions
,
1996,
ArXiv.
[6]
Amos Fiat,et al.
On-line load balancing with applications to machine scheduling and virtual circuit routing
,
1993,
STOC.
[7]
David S. Johnson,et al.
Computers and Intractability: A Guide to the Theory of NP-Completeness
,
1978
.
[8]
Keith A. Bartlett,et al.
A note on reliable full-duplex transmission over half-duplex links
,
1969,
Commun. ACM.