page lookups slower when use css11506?!

[julxu]

I have configured one armed test load balancing for L4 (http) as below:

service web-test1

ip address 10.10.10.253

keepalive type tcp

keepalive port 8100

active

service test2

ip address 10.10.10.254

keepalive type tcp

keepalive port 8100

active

owner WEB-Farm

content http-test

vip address 10.10.1.56

add service web-test1

add service web-test2

balance aca

port 8100

protocol tcp

active

group http-test-group

vip address 10.10.1.56

add destination service web-test1

add destination service web-test2

active

Our test result shows it slow than just software load balancing on http server.

Please confirm if my configuration is right?

If so, is one arm configuration will slow the lookup speed?

if not, what is the problem can be? and how to deb the problem?

Any comment will be appreciated

Thanks in advance

[smahbub]

The configuration part looks good. With this configuration the one arm load balancing should work fine. If you are running into slow connectivity issues, then there must some other thing that is causing the problem. Is this a new setup or is an old setup. ?

[julxu]

Great thanks for the reply.

it is a new setupped test environment.

The test staff said:

I did notice some strange behaviour that may/may not be related to this problem. The test going through the CISCO switch took a very long time to 'ramp-up' it processing - i.e. it was about 25-30 minutes of basically very slow processing and then later on it sped up.

Is possible I need change some option?

Is possible there is spanning tree problem?

It is not routing problem because the box is configured static routing ( all to next hop - cisco route)

It hardly to belive it is web server problem, because the test use same servers for both tests, that is load balance thought css11506 and thought a software resident inside one of web server.

Why it is slow at first half hour and than it sped up?

[Gilles Dufour]

could you give it a try with 'balance leastconn' instead of balance aca.

The later is very particular and may give strange results in a testing environment.

I would not recommend it unless you know exactly what kind of behavior to expect when using it.

Regards,

Gilles.