Re: Testprogram

To: " users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: Testprogram
From: Jeremy Kerr <jeremy@xxxxxxxxxxxxxxxxxxxxxx>
Date: Sun, 23 Feb 2003 19:11:49 +0800
> Does anybody now of a testprogram that can simulate traffic from lots of
> different ipaddresses?

Julian Anastasov's testlvs program will generate traffic from a large number 
of source IP addresses. The packets are logged and dropped at the realserver, 
so no connections are actually established. Good if you're testing the 
director, but this won't give any indication of the performance of the server 
cluster; I'm guessing that you wish to test the output of the whole cluster 
here, not just the director...

I've been using httperf for benchmarking  - It doesn't generate traffic from 
different addresses, but if you do separate testing with both httperf and 
testlvs, you should get the data you need.

> The lvs does
> its job here but when we have such a small number of ipaddresses/clients it
> does not balance evenly.

This is odd; I've done a bit of testing from a small number of IP addresses, 
and the request load was evenly distributed amongst the servers precisely.

How do you mean that the load 'does not balance evenly' - are the load values 
on the servers drastically different? or are you looking at the ActiveConns 
data in ipvsadm -L ? If it's the former, could you describe the cluster? 
(filesystem replication?, identical machines?, etc)

The actual allocation of connections will depend on the scheduler - which one 
are you using? (rr/wrr/lc/wlc/etc?). 

> We'd like to ba able to specify wich files to get with either ftp or http
> to simulate real downloads from files actually on the storage. It would be
> nice also to be able to specify a timeout when the testprogram thinks the
> realserver is inaccessable.

httperf will do both - it is similar to apachebench, but allows finer control 
over the request load and gives lots of detail about the response 
characteristics. you can get httperf from

O'Rourke and Keefe have written a paper on benchmarking LVS - . It provides a good 
amount of detail about how they performed their tests.


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