directory-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alex Karasulu" <akaras...@apache.org>
Subject Re: Benchmarks feedback
Date Tue, 27 Mar 2007 20:25:20 GMT
On 3/27/07, Quanah Gibson-Mount <quanah@stanford.edu> wrote:
>
> --On Tuesday, March 27, 2007 4:04 PM -0400 Alex Karasulu
> <akarasulu@apache.org> wrote:
>
> >
> >
> > On 3/27/07, Quanah Gibson-Mount <quanah@stanford.edu> wrote:
> >
> > Java based load generators are just fine.  I guess this is a tactic to
> > toot the OpenLDAP (C) horn
> > and bash Java while doing it: subtle but apparent.  Even Java clients
> are
> > not good
> > enough I guess :/.
>
> No, that wasn't the intent.  The fact is, a single C based client can
> generate 25 times more authentications/second than a single slamd client.
> So, that would mean I could use 75% fewer clients if I could use C, to
> generate a similar load.


Depends on the client's implementation!  You cannot make a general comment
like this.

However if you say it seriously enough like a CNN anchor or something people
can believe
anything right?



> > The whole point to load generators is to increase the number of them as
> > needed to
> > blast the server with as much load to saturate it.  If one load injector
> > is not sufficient then
> > increase the number.  SLAMD does the job nicely.
>
> Well, actually, it is a little more than that.  The more clients you put
> on
> a single system, the worse things get.  I currently use 5 systems with 4
> clients each, for a total of 20 clients.  Fewer servers with more clients
> performs worse.



The point was you increase the number of client's (machines) rather than
processes
that are hitting the server.  Again these far fetched conclusions depend on
the
SLAMD tests and how they were written.

I've seen Java code outperform C code and vice versa.  It all depends on how
it was
written doesn't it?

Alex

Mime
View raw message