Often channel bonding does not bond fully. Ie, with an iperf test, you would need at four connections, one using each channel, to use the full bandwidth.
-----Original Message-----
From: GRIDPP2: Deployment and support of SRM and local storage management [mailto:[log in to unmask]] On Behalf Of Ben Waugh
Sent: 18 January 2011 15:28
To: [log in to unmask]
Subject: Fwd: Channel bonding throughput tests
Hi All,
This is probably a better list for this than the LondonGrid technical
list I tried first...
I am setting up channel bonding using 4 1-Gb/s NIC ports on each of our
two latest disk servers. This is now working, after the UCL network
group fixed a configuration error on the switch. However, a naive iperf
test does not show any difference in the throughput between the two
machines. Should I expect a difference, or is a more sophisticated
approach required to make use of the wider link?
Does anyone have other suggestions, for example for the bonding mode to
use. I've tried all the ones that load-balance, with no obvious
difference on the iperf results.
Cheers,
Ben
--
Dr Ben Waugh Tel. +44 (0)20 7679 7223
Dept of Physics and Astronomy Internal: 37223
University College London
London WC1E 6BT
|