Bandwidth for High Speed Networks

Note that as long as all three ows attempt to use

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

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: cket will re ect the new rate of the ow. To test how well this works in practice, we consider the topology in Figure 8, where each link is 10 Mbps. Note that as long as all three ows attempt to use their full fair share, the fair shares of ows 1 and 2 are less on link 2 3.33 Mbps than on link 1 5 Mbps, so there will be dropping on both links. This will test the relabelling function to make sure that the incoming rates are accurately re ected on the second link. We perform two experiments only looking at CSFQ's performance. In the rst, there are three UDPs sending data Sources Flow 2 1 (10 Mbps) 10 Mbps 10 Mbps Gateway Flow 1 2 (10 Mbps) (Link 1) 10 Mbps Gateway (Link 2) 10 Mbps Sink 10 Mbps 8 , and have bene ted greatly from conversations with Steve Deering and Sally Floyd. We should note that the matters addressed in this section are rather controversial and this overview unavoidably re ects our prejudices. This section, however, is merely intended to provide some perspective on our motivation for this work, and any biases in this overview should not undercut the technical aspects of the CSFQ proposal that are the main focus of the previous sections. 4.1 The Unfriendly Flow Problem Flow 3 (10 Mbps) Figure 8: Simulation scenario for the packet relabeling experiment. Each link has 10 Mbps capacity, and a propagation delay of 1 ms. Tra c Flow 1 Flow 2 Flow 3 UDP 3.36 3.32 3.28 TCP 3.43 3.13 3.43 Table 4: The throughputs resulting from CSFQ averaged over 10 seconds for the three ows in Figure 8 along link 2. at 10 Mbps each. Table 4 shows the average throughputs over 10 sec of the three UDP ows. As expected these rates are closed to 3.33 Mbps. In the second experiment, we replace the three UDPs by three TCPs. Again, despite the TCP burstiness which may negatively a ect the rate estimation and relabeling accuracy, each TCP gets close to its fair share. 3.7 Discussion of Simulation Results We have tested CSFQ under a wide range of conditions, conditions purposely designed to stress its ability to achieve fair allocations. These tests, and the others we have run but cannot s...
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