|
Edward Chow wrote:
>
> Joseph,
>
> Will modify WebStone benchmark tool, which is multiclient (webclients), with
> testlvs functions address the problem?
> This can be a good student semester project.
I don't know. There are a lot of possibilities here and it's not obvious where
to go.
If you're going to commit someone to a semester's work,
it would be worth some time figuring out what to do,
so that at the end, when the student is applying for jobs,
they can point to their code being used in the real world.
Otherwise there will just be another program lying around
that no-one is using.
Polygraph is the "all things to all men" package. It is being actively
developed and will be for as long as the developers continue to be funded.
It is accepted as a benchmark by commercial providers of webhard/software,
but there really isn't any competition in this area, so we don't have
a lot of options here. It's also possible that polygraph is so well suited to
it's job that there is no point in writing another program.
When you use webstone as a multiclient, do the connections all go to the
same real-server or to different real-servers, ie does it appear as a
multiclient
to LVS?
I haven't used testlvs and I don't know enough to say whetheer
anything is gained by making it multi-client.
I think it just beats up the director and it seems to do that
quite well from what I hear on the list - it may not need to be multiclient.
My favorite stress tester is netpipe (see the performance page on
the LVS website). The netpipe author is working on a multi connect
version (but not very actively when I last contacted him - early 2000)
and he'll give you the preliminary code if you ask him (well
he gave it to me, and all I did was ask). netpipe will happily saturate
your network bringing a VS-NAT director to its knees (but not a VS-DR
director). I can't do that with any of the other tools I have.
Netpipe works in two modes:
sends a packet of user chosen size and receives in return the same packet
sends packets of user chosen size and receives in return an <ack> ie a small
packet.
It would be nice if netpipe could send a trigger packet and receive a reply of
user chosen size.
You could make netpipe better, by adding multi ping or testlvs to it if
you like.
So I have two thoughts with my limited understanding of the matter.
- see if there's anything you can add to polygraph that would test LVS.
- see if the netpipe author is interested in you finishing the multiconnect
version of netpipe
Joe
--
Joseph Mack PhD, Senior Systems Engineer, Lockheed Martin
contractor to the National Environmental Supercomputer Center,
mailto:mack.joseph@xxxxxxx ph# 919-541-0007, RTP, NC, USA
|
