On Thu, Jan 27, 2011 at 04:22:49AM -0800, Joseph Mack NA3T wrote:
> On Thu, 27 Jan 2011, Malcolm Turnbull wrote:
>
> > Simon,
> >
> > We've just ordered 4 servers and a gigabit switch to try and do exctly
> > this kind of testing.
> > If anyone has sugestions about a good/simple way of generating the
> > load and graphing it that would be appreciated.
>
> I used netpipe back in 2000
>
> http://www.austintek.com/LVS/performance/
>
> The only problem with netpipe is that you can only setup one
> connection at a time. This was fine to test the
> director/realserver to its network limit, but you can't test
> LVS running multiple connections
Hi Malcolm, Hi Joe,
The last time that I did some micro-benchmarking of LVS I found Julian's
lvs-test to be of use. http://www.ssi.bg/~ja/
More generally, when doing network benchmarking I find netperf to be very
useful. Though most of my work using netperf has been for single streams,
which is not particularly exciting in the context of LVS.
The in-kernel packet generator is also interesting, though I don't recall
ever getting particularly useful benchmarks out of using it.
_______________________________________________
Please read the documentation before posting - it's available at:
http://www.linuxvirtualserver.org/
LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
or go to http://lists.graemef.net/mailman/listinfo/lvs-users
|