CSS BEA JSESSIONID delay issue

[jfoerster]

Hi,

my customer experiences strange delay variantions from accessing the webserver directly instead of accessing the servers via the CSS.

the CSS is configured for cookiestickiness with the string commands.

The difference in delay is as following:

Accessing the server directly: 0.2 seconds until the page is delivered

Acessing the service via the VIP 0.4 seconds until the page is delivered.

this factor 2 can be seen everytime we are accessing the Bea-Webpage. If a certain request takes 0.7 seconds accessing the servers directly we have approximately 1.4 seconds via the CSS.

Turning back to a Layer3 rule is right now out of the game nevertheless I try to arrange a testing window where I'll give it a try using a layer3 rule and src-ip-stickiness instead of a layer5 rule with cookie(JSessionID) stickiness.

Any idea what might be the cause of this delay and how to solve this issue is appreciated.

Kind Regards,

Joerg

[Gilles Dufour]

Joerg,

capture a trace when going directly and when going through CSS. Then compare delay between packets.

Take a trace frontend and backend as well.

Try also longer donwload - less than 10sec times are difficult to compare.

Adding a device switch or router will always introduce a delay.

Also, are you comparing just a single GET request ?

The CSS in L7 mode will switch the first request in software and then move the service to fastpath.

So the first request will always take longer but subsequent requests inside the same tcp connection should go faster.

Finally, when comparing the CSS to a server you will most probably see better performance on the server with a small amount of connections.

Give it another try with 50k connections or more and now I'm pretty sure the CSS will look better than a single server.

I'm sorry to say that the current measuring is not very good to validate a performance issue.

Gilles.

[jfoerster]

Gilles,

just to update you.

We investigated the inssue and found out that some database queries are delaying the webtraffic. Nevertheless we experience a lot of diffrences in terms of delay if we access the server directly or via the loadbalancer. But so far this is issue which we handed over to the server guys. So we are awaiting their feedback.

If any issue arises again I'll get back to this posting.

Cheers,

Joerg