Bandwidth for High Speed Networks

Certainly csfq is far superior in this regard to the

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: how here because of space limitations, suggest that CSFQ achieves a reasonable approximation of fair bandwidth allocations in most conditions. Certainly CSFQ is far superior in this regard to the status quo FIFO or RED. Moreover, in all situations CSFQ is roughly comparable with FRED, and in some cases it achieves signi cantly fairer allocations. Recall that FRED requires per-packet ow classi cation while CSFQ does not, so we are achieving these levels of fairness in a more scalable manner. However, there is clearly room for improvement in CSFQ, as there are cases where its performance is signi cantly below that of its benchmark, DRR. We do not yet know if these are due to our particular choices for the estimation algorithms, or are inherent properties of the CSFQ architecture. 4 Why Are Fair Allocations Important? In the Introduction we stated that one of the underlying assumptions of this work is that fairly allocating bandwidth was bene cial, and perhaps even crucial, for congestion control. In this section we motivate the role of fair allocations in congestion control by discussing the problem of unfriendly ows, and then presenting two approaches to this problem; we end this section with a discussion of the role of punishment. In what follows we borrow heavily from 7 , 3 , and Data networks such as the Internet, because of their reliance on statistical multiplexing, must provide some mechanism to control congestion. The current Internet, which has mostly FIFO queueing and drop-tail mechanisms in its routers, relies on end-to-end congestion control in which hosts curtail their transmission rates when they detect that the network is congested. The most widely utilized form of end-to-end congestion control is that embodied in TCP 11 , which has been tremendously successful in preventing congestion collapse. The e cacy of this approach depends on two fundamental assumptions: 1 all or almost all ows are cooperative in that they implement congestion control algor...
View Full Document

This document was uploaded on 03/04/2014 for the course ENG 531 at Rice.

Ask a homework question - tutors are online