LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: LVS Performance Testing

To: DongHyun Kim <ryan@xxxxxxxxxxxxxxxxx>, lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Subject: Re: LVS Performance Testing
From: Drew Streib <ds@xxxxxxxxxxx>
Date: Sun, 11 Jun 2000 09:52:14 -0700
On Sun, Jun 11, 2000 at 03:31:31PM +0900, DongHyun Kim wrote:
> > 
> > > COuld you please
> > > 1. suggest a better traffic generator than Apache Bench.. 
> > 
> > still not a solved problem. 
> > 
> 
> how about http_load ?

The only way to really simulate real-world traffic would be to alter
lag times, have percentages of simply dropped tcp connections, malformed
packets, etc.

The only way to do this right is to alter the testing machine's tcp stack
behavior in the kernel, or run the test through some specialized hardware
designed to introduce these abnormalities (this hardware does exist).

Nistnet attempts to do this in software and is the best i've seen so
far at this. It involves a linux kernel patch, however.

The short of it is, that real world testing is either very 
difficult (impossible with a tool like ab) or very easy (put the
machine in actual real-world production). :) The only way I would trust
someone's lab network benchmarks is if they could show a correlation
between their benchmark loads and real-world loads on production machines.
This is something I'm working on right now. 

-drew

-- 
-------
Drew Streib <d@xxxxxxxxxxx> 408.542.5725

Technical Marketing Manager, VA Linux Systems | <dtype@xxxxxxxxxxx>
Sr Developer, Community Liason, SourceForge   | <dtype@xxxxxxxxxxxxxxx>
System Administrator, Linux International     | <dtype@xxxxxx>


<Prev in Thread] Current Thread [Next in Thread>