Ftp trace data from the Fairisle ATM network

This directory contains several data files which individually contain measurements from ftp and other traffic transferred over the Fairisle network. The measurements were made using code written by Simon Crosby, and make use of the ATM infrastructure developed by members of the Systems Research Group at the University of Cambridge.

This data is Copyright (c) Simon Crosby, 1994, University of Cambridge, UK. You may freely distribute this data but must retain the copyright notice.


Details of the Fairisle network and the Systems Research Group activities can be obtained from the ftp site at Cambridge:

ftp host is:	ftp.cl.cam.ac.uk
login:		anonymous
directory:	reports/ATM/docs-94-3/*


This is also available as the Systems Research Group Technical Note of March 1994 (the blue book), and can be obtained from the technical reports service at Cambridge: send email to techreports@cl.cam.ac.uk


ftp host is:	ftp.cl.cam.ac.uk
login:		anonymous
directory:	fairdata/ftp-trace/
files:		as described above		- trace data
		readme				- this note



Experimental data

Currently the data is minimal: three experiments were run, each of which yielded 3 files of data. All files within an experiment have the same base name e.g. ftp-fast-bernoulli. More complex experiments are in progress.

The topology for the first experiments was as follows:

Unix ftp server + ATM <--> switch 1 <--> switch 2 <--> Unix ftp client + ATM
The ftp client requests data (the data varies, see below) from the ftp server over the Fairisle network. Both client and server interface directly to the ATM network. Both of these machines are DECStation 25000 mips machines running Ultrix 4.3. The ATM protocol stack and connection management are described in the blue book. The ATM switches are Fairisle fpc3-based 8x8 switches. Each host was equipped with an ORL Yes V2.2 ATM interface (described in the blue book). This interface has limited buffer space and operates in polled I/O mode; consequently its performance is severely limited.

Within an experiment <name> the files are:

File format

Each file is in ASCII format, and each line (i>0) is a single measurement. Each measurement (i) records the time, in ticks, between the absolute cell number (i) and absolute cell number (i-1). Measurement (1) is therefore meaningless.

A tick is a fabric cell time. The fabric cell time in these experiments was set to 3.56 microseconds. The transmission line rate was set to 4.4 microseconds per cell. A pair of back to back cells is recorded as an inter-arrival time of 1 tick.



This records the cell trace for a single ftp of a directory containing a single large file (the operating system image), with no other multiplexed traffic through the switch. This is thus the cell trace for the ftp process only. The ftp used binary mode transfer and the data was copied directly to /dev/null, increasing the throughput. Note that the throughput is low because we were using an early design of the ATM host interface.


This records the behavior of the multiplex of the data described above with 6 independent sources each of which transmits a bernoulli traffic trace with a mean rate of about 1/32 of the fabric bandwidth. Each source originates at a different input port on switch 1 and is routed through the multiplex with the ftp data to switch 2 where each connection is terminated at a different sink (the idle input ports of switch 2)


This records the cell trace for a single ftp of a directory containing a large number of postscript files (the ftp directory of the blue book). The data was saved on a networked file system, and therefore the performance is much lower. This traffic was multiplexed with the bernoulli traffic described above.