Links to Software and Datasets

1995 WWW Client Datasets

These datasets are the basis for many published studies. They are available from the Internet Traffic Archives. The format of the traces and collection process are documented in our tech report BUCS-TR-1995-010.

1998 WWW Client Datasets

In 1998 we captured a new set of client logs, using a method different from the 1995 set. The format of the trace data and collection process are documented in our tech report BUCS-TR-1999-011, and the trace itself is here.

tcpeval

Tcpeval constructs critical path analyses of TCP transactions. It was developed by Paul Barford during his PhD thesis research. Its algorithms are described in the paper Critical Path Analysis of TCP Transactions in Proceedings of the 2000 ACM SIGCOMM Conference, Stockholm. Sweden, September 2000.

The source code is available in a compressed tarfile here. Included in the tarfile is a HOWTO with installation instructions.

If you download this code, please send an email to Paul Barford (pb at cs.wisc.edu) let him know you are using it, and whether you find it useful.

Surge

Surge, which generates Web requests intended to mimic measured statistical properties is availble here.

The paper describing Surge's rationale and design is Generating Representative Web Workloads for Network and Server Performance Evaluation in Proceedings of Performance '98/ACM SIGMETRICS '98.

However, the default models and parameter settings used in this version of Surge are based on analyses of the 1998 dataset, documented in Changes in Web Client Access Patterns: Characteristics and Caching Implications in World Wide Web, Special Issue on Characterization and Performance Evaluation, Vol. 2, pp. 15-28, 1999.

This is the HTTP/1.1 compliant version of the code (HTTP/1.0 is still supported in this release). There is a detailed HOW-TO included which should get you going.

If you download this code, please send an email to Paul Barford (pb at cs.wisc.edu) who developed it and will put your name on the SURGE interest mailing list so that you will be notified about future updates. We'd also be interested in what you will be using the code for - if you could give him a brief overview I would appreciate it.

While the HOW-TO suggests using the MIT pthreads, if you are using a 2.2 Linux kernel, we recommend you compile using kernel threads (make sure your thread limit is set high enough!). To do that make the following mods:

In the Makefile, replace PGCC by CC. Added the pthread library to Surgeclient.c rule (ie. -lpthread and -D_REENTRANT).
Remove pthread_init() from Surgeclient.c
Added #include <netinet/in.h> to Surgeclient.c (it is needed for the struct sockaddr_in). (For Linux this should be #include <linux/in.h>).
In Surgeclient.c you may need to change sys/times.h to sys/time.h (no plural)

BPROBE

BPROBE is a tool for measuring bottleneck bandwidth of an Internet path, using the packet-pair technique. It was developed by Bob Carter (carter at cs.bu.edu) during his PhD research. Source code for BPROBE is available here and the paper describing the design of BPROBE is here.

Aest: A Tool For Estimating the Heavy Tail Index from Scaling Properties

This tool provides an estimation of the tail index alpha for empirical heavy-tailed distributions, such as have been encountered in telecommunication systems. It uses a method (called the ``scaling estimator'') based on the scaling properties of sums of heavy-tailed random variables. The software is available here, and the paper describing aest is available here.

All code on this page is licensed under a Creative Commons License.

Mark Crovella /