Report on individual activity to the PPDG for the quarter ending July 31, 2001
 
Les Cottrell, SLAC

Project: Network throughput performance

We have been making measurements of bulk network throughput between SLAC 
and major BaBar sites using iperf. These measurements
are reported in http://www-iepm.slac.stanford.edu/monitoring/bulk/. 
The goal is to understand how to achieve high performance and the impact 
of large windows and multiple parallel streams 
(see http://www-iepm.slac.stanford.edu/monitoring/bulk/window-vs-streams.html). 
From this we can set expectations of what is achievable and how to go about 
achieving it. We are also working  to
understand what is the likely impact on others using the links at the same time. 
We have also worked on validating how well simulation
can represent the iperf meaurements (see 
http://www-iepm.slac.stanford.edu/monitoring/bulk/ns.html). 
If successful this will simplify and reduce the amount of effort in 
making measurements. It also opens
up the door to in-line predictions and dynamic adjustment of TCP etc. 
parameters by applications to improve throughput.

Recently we have extended the measurements to the application level, 
and are using the SLAC written bbcp file copy program to make measurements
of file transfer performance between sites. The goals of this are to 
see what extra constraints are imposed by the application on top of the
network layer (e.g. security, disk access etc.), and how close one can 
get to network (iperf) performance. The early results are available at 
http://www-iepm.slac.stanford.edu/monitoring/bulk/bbcp.html. 
We hope to present more complete information at CHEP01.

We are also starting to look into how we can use QBone Scavenger Service (QBSS) 
to reduce the impact of the high throughput on other
users. We will be reporting on this at the Internet 2 meeting in Austin, 
in October 2001.