Pchar network tool

In Summer , I wrote the data processing part of clink, which became one of the source code files below process. I used the alpha version of pathchar to collect data, and wrote a paper describing my experiences, and proposing some improvements to pathchar's data processing. In Summer , I wrote the data collection part of clink, which is the source code file called collect. So far, the primary differences between pathchar and clink are: 1 clink uses the even-odd technique described in the SIGCOMM paper to generate interval estimates for bandwidth.

The source code for clink is available here in a tar file 82 KB or at gzipped tar file 23 KB. It is also an Open Source reimplementation of pathchar, available from Bruce's web page.

Abstract We evaluate pathchar, a tool that infers the characteristics of links along an Internet path latency, bandwidth, queue delays.

Looking at two example paths, we identify circumstances where pathchar is likely to succeed, and develop techniques to improve the accuracy of pathchar's estimates and reduce the time it takes to generate them. Introduction pathchar is a new tool, written by Van Jacobson at Lawrence Berkeley Laboratory LBL , that tries to infer the characteristics of individual links along an Internet path by measuring the round trip time of packets sent from a single host.

We explain the basic mechanism and evaluate its accuracy on two paths whose link characteristics are known. Based on these observations, we propose techniques to improve the accuracy of pathchar and to reduce the number of measurements and time it takes to generate its estimates. The contributions of this paper are An evaluation of pathchar and some insight into when it can or cannot be expected to be useful.

A technique for generating intervals for the estimates pathchar generates. A technique for determining dynamically the number of measurements needed to achieve a given accuracy. Conclusions Based on our evaluation of the alpha version of pathchar we conclude: Estimating link latencies is relatively easy, since latencies in wide-area networks are large compared to pathchar's measurement errors.

Estimating bandwidths is harder, because the difference in round-trip time between the largest packet and the smallest is small compared to the measurement errors.

The higher the bandwidth, the more difficult it is to estimate. The key to getting a good bandwidth estimate is to send enough probes that one of them traverses the entire path without incurring any queue delays. As the length of the path increases, the number of probes required increases quickly.

A single busy link, by imposing queue delays on the majority of probes, makes it difficult to resolve the characteristics of links on the other side.

On the other hand, slow links are not necessarily a barrier to accurate measurement, as long as their performance is consistent. Of the new techniques we tested, only adaptive data collection seems to work well: it greatly reduces the amount of data required, without affecting the accuracy of the estimated characteristics.

Principally contains build fixes for Linux and other minor bugfixes. This is likely to be the last release of pchar. New features include multi-packet probes inspired by LBL's pipechar utility , kernel-level timestamps via the pcap library, and TCP probe packets. The current version of pchar is 1. It is available as a gzipped tar file. This distribution is signed with the following PGP public key, available from one of the author's Web pages or from most PGP keyservers:.

There have been reports of success with pchar on: FreeBSD 2. Allen B.