Forum Discussion

pjcampbell_7243's avatar
May 09, 2013

Very slow going through LTM vs direct to host

Hey all...

 

I see a huge lag when measuring download times for a static or dynamic file through the LTM using Chrome's "network" tab under developer tools.

 

An example is 2.5 seconds for a 700KB file directly to the host, and 13+ through the F5.

 

 

In trying to diagnose the problem, I've setup a test VIP (HTTP with HTTP profile) and a single Linux Apache server pool member.

 

 

What I can't understand is I just do not see the delays in curl. What should this be telling me?

 

Of importance: from reading other posts, I've tried changing the tcp profile from "tcp" to "tcp-lan-optimized" and this seems to speed things up by quite a bit. "tcp-wan-optimized" seems faster still, in this particular case. It's actually measuring faster than going directly to the host, which I'm not sure I understand.

 

Further updates, one some sites, using the lan profile seems to make the site significantly slower (?). I've also read the suggestion of disabling nagle, which is enabled by default on both optimized profiles.

 

I have 610 VIPs so my question is how do I know which to change or which to leave alone? Most of these apps are used by wan users, with some users coming from the internet and some over a metro ethernet connection to the app.

 

Unit is a 1500 LTM with 10.2.4 HF6

 

2 Replies

  • Have you tried using alternative tools like IEWatch with IE and HTTPWatch with FF? Just to see if the results are similar?

     

     

    It's probably worth doing a tpcdump on the F5 to get a better idea of the delay regardless.

     

     

    Also, where the tcp profiles are concerned I'd suggest you don't use Nagle server-side at least and enable delayed acks instead.
  • I've run into this a few times and if you can live without having to parse http (e.g. no irules, no http profile, no cookies, etc), you can go with a performance type virtual server and adding a fastl4 profile instead.