Scale / Throughput testing

[Brian DeLacey]

Kurt and did some testing over the weekend with Netpipe and a half-dozen
computers configured in a cluster. We've also done some testing with a
smaller configuration of three standalone machines.

We're working out the bugs in our testing scaffolding and learning about
some of the subtle nuances while ramping this up. (For instance, some radios
default to 1 Mbps even though they have a max speed of 54 Mbps. Is that
considered a bug in the Linux driver?)

So far, we have obtained varying results. One configuration represented our
"low", with Netpipe peaks in the 6 - 7 Mbps range. Perhaps a configuration
glitch? Another configuration had peaks in the 13 - 14 Mbps range. (We were
using the same radios in both tests.) These tests had no intermediate hops.
We also tested blocking one peer link and forcing a hop, with the resulting
peak cut in half - as expected?)

Netpipe is easy to run and readily accessible through a standard apt-get
install on Ubuntu. (A paper
here<http://www.scl.ameslab.gov/netpipe/paper/full.html>discusses the
tool.) It's easy to run on multiple machines at the same time
etc.

What are other people seeing for throughput numbers? Are there any
configuration recommendations? Once we get to a steady state with our
staging, we plan to scale the number of machines in the mesh test.

Brian
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://open80211s.com/pipermail/devel/attachments/20090323/20b11fd7/attachment.html