Re: http reqs. to 4 real servers does not scale

To: " users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: http reqs. to 4 real servers does not scale
From: "Stephan Hendl" <Stephan.Hendl@xxxxxxxxxxxxxxxxxx>
Date: Mon, 26 Feb 2007 14:31:18 +0100

>>> Graeme Fowler <graeme@xxxxxxxxxxx> schrieb am 26.02.2007 um 14:05 in 
>>> Nachricht
> On 26/02/2007 12:50, Stephan Hendl wrote:
> <snip>
>> Are there any kernel parameter that I can tune? Or other ideas?
> Several questions for you, first:
> Are you using persistence?

the default lvs parameters only

> What tool are you using to run the test?
http_load (

root# ./http_load -parallel 200 -seconds 60 ../ism_baseline.txt

> Is your filesystem blocking the requests in some way?
no it isn't. There are no relevant entries in the /var/log/messages. The vmstat 
io sections show values nearly zero because of the samll files.

> Do you get any errors in your Apache logs (assuming Apache server)?
no error messages at all

> Is your test client capable of generating more than 9500 reqs/sec? I've 
> (one more than one occasion) caused a test client running Apachebench to 
> fall over because I've exhausted the test client's resources long before 
> those of the cluster.
I don't know exactly. It is a hp proliant dl380g2, 2x 1.4 GHz with 100Mbit LAN.

I tried the same procedure with two servers, the second is a more modern one 
with 2x 3 GHz. The results are:

server 1: 
484837 fetches, 200 max parallel, 2.13328e+07 bytes, in 60 seconds
44 mean bytes/connection
8080.62 fetches/sec, 355547 bytes/sec
msecs/connect: 11.4204 mean, 8994.39 max, 0.226 min
msecs/first-response: 11.0906 mean, 829.172 max, 0.479 min

server 2:
526187 fetches, 200 max parallel, 2.31522e+07 bytes, in 60 seconds
44 mean bytes/connection
8769.78 fetches/sec, 385870 bytes/sec
msecs/connect: 11.0908 mean, 9007.96 max, 0.233 min
msecs/first-response: 10.2539 mean, 1486.05 max, 0.424 min

no, with two server at the same time I got the apache error:

[Mon Feb 26 12:47:27 2007] [error] server reached MaxClients setting, consider 
raising the MaxClients setting

So it might be that one server cannat handle more than 9500 req/s! What do you 


> Graeme
> _______________________________________________
> mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx 
> Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx 
> or go to

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