Loading...
Please wait, while we are loading the content...
Similar Documents
Tcp over End-to-end Abr: Tcp Performance with End-to-end Rate Control and Stochastic Available Capacity
| Content Provider | Semantic Scholar |
|---|---|
| Author | Shakkottaiy, Sanjay Kumarz, Anurag Karnik, Aditya |
| Copyright Year | 2007 |
| Abstract | We consider TCP over the Available Bit Rate (ABR) service in an ATM network , when sharing the network bandwidth with time-varying CBR/VBR traac. It is assumed that the TCP end-points are also ATM/ABR end-points. We develop an analysis, using which we compare TCP throughput, with and without end-to-end ATM/ABR transport. The TCP connection is assumed to be bottlenecked at a single link in a wide-area network; the capacity of this link available to the TCP connection is time varying (owing, e.g., to being shared by guaranteed service traac). We also provide results from a simulation on a TCP test-bed that uses actual Linux TCP code, and a simulation/emulation of the network model inside the Linux kernel. We show that, over ABR, the performance of TCP improves signiicantly if the network bottleneck bandwidth variations are slow as compared to the round-trip propagation delay. Further, we nd that TCP over ABR is relatively insensitive to bottleneck buuer size. These results are for a short term average link capacity feedback at the ABR level (instantaneous capacity (INSTCAP)). We use the test-bed to study another rate feedback based on a longer term history of the capacity process. We call this EFFCAP feedback, as it is motivated by the notion of the eeective capacity of the bottleneck link. EFFCAP feeds back the minimum over several (a parameter N) short term averages (averaging interval set by a parameter M). We nd that EFFCAP feedback is adaptive to Research supported by a grant from Nortel; paper submitted to the IEEE Transactions on Networking y Currently with the 1 the rate of bandwidth variations at the bottleneck link, and thus yields good performance (as compared to INSTCAP) over a wide range of the rate of bottleneck bandwidth variation. We then study the eeect of the choice of M and N on the throughput performance of TCP. We provide a guideline for choosing values of these parameters that work for a wide range of the rate of bottleneck bandwidth variation, and the session round trip time. Finally, we consider the model with two TCP connections to study if TCP over ABR provides throughput fairness even if the two connections have diierent round-trip propagation delays. We study the choice of M for the two connections case; in particular, we study whether diierent values of M should be used for EFFCAP feedbacks to the two connections, or whether an average … |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://www.comm.csl.uiuc.edu/~shakkott/Pubs/ton.ps.gz |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
