From: rick.jones2@hp.com
Morten Reistad wrote:
> In article ,
> Rick Jones wrote:
> >Jorgen Grahn wrote:
> >> - Are you allowed to put extra load on the server, stealing its CPU
> >> resources or SIGSTOP the server software momentarily?
> >
> >Might be OK for simulating a very heavily CPU loaded server, but
> >I'd avoid it for simulating a heavily loaded network. I was
> >involved in a benchmark that was doing something conceptually
> >similar - not calling recv() except at a rate based on the presumed
> >simulated bandwidth - and the zero window situations were inducing
> >TCP behaviours a realy link of that bandwidth would not have
> >induced. I'm reasonably confident that SIGSTOP would do the same
> >thing.
> If you are on Linux you can use tc(8) to simulate all kinds of
> networks; you can make drops, jitter, duplicates, delay etc; just
> as a real network will.
Yep, and I think *BSD has Dummynet, and there is an unofficial, crude
and now aging port of dummynet to HP-UX. I suspect there is something
along those lines for Solaris, perhaps even AIX and I wouldn't be
surprised if there were something for Windows. So, there should be
plenty of opportunity to simulate a server located in far far away
places.
When I was doing things the the HP-UX port of dummynet, I tried to
keep it off the System Under Test (SUT) to keep it from loading the
CPU. I'd probably try to do the same thing here unless the network
being simulated is so slow that the CPU horsepower on the SUT is
massively more than required for the regular software. Even if just
out of unfounded paranoia about simulation overheads :)
rick jones
--
portable adj, code that compiles under more than one compiler
these opinions are mine, all mine; HP might not want them anyway... :)
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...
--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)
|
