DRAFT VERSION The authors are still struggling with the open problem presented in Section 3. Please let us know if you are working on it. Abstract By viewing the popular TCP (Transport Control Protocol) as a distributed scheduling algorithm , we consider the competitiveness of it against the optimal global algorithm in minimizing the average \user perceived latency" or \\ow time" of the jobs. The paper by Edmonds, Datta, and Dymond solves the problem for the single bottleneck case. TCP (Transport Control Protocol) can be much more complicated when jobs pass though a general network. With extreme parameter values, TCP will behave smoothly without the usual causes of chaos, however, a steady state is still not guaranteed. Modulo this diiculty, we are able to prove that on a network in which each transmission passes through at most m bottlenecks, TCP with O(m 4) times as much bandwidth is O(1) competitive, against the optimal global algorithm. 1 Non-Introduction This paper as of yet does not have a proper introduction that motivates TCP and the problem and that outlines the previous results. There are two reasons for this. The rst is that the authors are still struggling with the open problem presented in Section 3. Though the paper may be publishable the way it is, it would be much better to have this problem solved. The second reason is that the introduction would be very similar to that for the paper by Edmonds, Datta, and Dymond EDD01] that solves the problem for the single bottleneck case. You may want to rst read the introduction for that paper. On the other hand, every thing you need is here. The general transmission network G is modeled as follows. There are a number of routers that act as bottlenecks. Each has a maximum bandwidth B k and \loses" random transmission beyond this maximum. The input consists of a set of jobs (or sessions) J = fJ i g to be scheduled. Each job J i is deened by its arrival time a i , its le length l i , and the subset of the bottlenecks B(i) that it passes through. Conversely let J t (k) denote the set of jobs J i that pass through the k th bottleneck and have not completed at time t. The graph specifying the connection of the transmission channels and the order that a job's transmission passes through …
[1]
Raj Jain,et al.
Analysis of the Increase and Decrease Algorithms for Congestion Avoidance in Computer Networks
,
1989,
Comput. Networks.
[2]
Kirk Pruhs,et al.
Broadcast scheduling: when fairness is fine
,
2002,
SODA '02.
[3]
Jeff Edmonds,et al.
Scheduling in the dark
,
1999,
STOC '99.
[4]
Sally Floyd,et al.
Connections with multiple congested gateways in packet-switched networks part 1: one-way traffic
,
1991,
CCRV.
[5]
Bala Kalyanasundaram,et al.
Speed is as powerful as clairvoyance
,
2000,
JACM.