Forum Discussion

Ev_28228's avatar
Icon for Nimbostratus rankNimbostratus
Nov 05, 2010

Is additional latency of 100% normal for an F5 Load Balancer??

Hi all,



We have just setup a production account with cloud infrastucture provider GoGrid (who provide free F5 LB as part of their package). What we notice is a linear, proportional delay in all our web round-trips when using the LB with an average delay of over 100% - which seems ridiculous to me but GoGrid are adamant that everything is correctly configured their side and that LB will "...of course add some latency to your requests...".



We have 1 F5 LB sitting in front of 2 web-servers (i.e. the simplest deployment possible).



Here's what I observe though Firebug or Chrome Dev Tools:


- Roundtrip to a particular page direct to the web-server: avg. 315ms


- Roundtrip to the same page through the LB: avg: 810s



Same test to a smaller page results in avg 350ms (without LB) and avg 680ms (with LB)





- No caching issues (all static content is cached) and payloads are identitical in both tests.


- We are using SSL - I asked GoGrid if perhaps the LB was also doing an SSL decryption/encryption and/or HTTP compression run (essentially duplicating what has already been done on the web-server) - They assured me this was not the case.


- These are steady-state results (i.e. not being skewed by first-time warm-ups / caching etc.)



Anyone have any ideas here as to what is going on (or how I can prove to GoGrid that something is wrong) - I sent them my screenshots of my timings but they can't explain it. I'm just trying to find out if this is normal. I would have thought that what is essentially a low-level, dedicated hardware switch would at most add a few milliseconds onto the round-trip time - not hundreds of milliseconds (with a load of 1 user!!).



Can anyone validate this? Any help would be much appreciated.










2 Replies

  • it really depends on the placement of the LB, but it's unlikely that it should introduce that much latency. Figure out what client TCP protocol profile they're using on your Virtual Servers. Ask them to switch it from tcp to "tcp-lan-optimized" and see if that helps.
  • I would have them give you a run down of your LTM configuration. It certainly does not sound like they have done it optimally.



    Trust but verify I always say... go look at some of the deployment guides that seem to fit your solution and compare it to what they say they have done.