Forum Discussion

Brent_J's avatar
Icon for Nimbostratus rankNimbostratus
Oct 13, 2010

Load Balancing K2 Blackpearl



Very new to F5 having only picked it up in the last 6 months and self learnt. Hoping that someone here will have already been thru this experience and be able to share their learnings with me. We've recently setup a load balanced K2 Blackpearl environment using F5 as the loadbalancers and are in the process of doing some performance testing. What we are finding is that the performance is terrible running through the F5s...


an example is that the avg time for one of the transactions bypassing F5 is 1.5 secs. When running via F5 it's 21secs.


Initially we didn't have the TCP Lan optimised profile applied and after applying that we saw a reduction in the timing to between 5 -10 secs for the same transaction which is still way too long. At the same time this is causing a flow on effect of loading the CPU on the K2host server to 27% cpu, where the same transaction load was using 10% CPU without F5 during the baseline tests.



We had one of our comms teams do a tcp dump on the traffic and his comment was there was a signicant number of TCP Window Size update requests when running via F5. We have tried altering the profile to turn off RFC1323 TCP Performance enhancements and this didn't appear to help things along any.



Trawling thru the documentation for Blackpearl there's no mention of Load Balancing other than the Microsoft NLB.


Question: Has anyone else successfully load Balanced K2 Blackpearl using F5 LTM? and if so was there any special settings configuration tweaks settings required to get the performance to acceptable levels.




FYI our Environment : BIG-IP 9.4.8 Build 396.1 Hotfix HF3



The test config is very simple 1 node, 1 port, no iRules, tcp lan optimised profile applied.




Thanks in Advance.





1 Reply

  • Brent: this is almost certainly a server-side issue. This is common in MS environments, particularly when NetBIOS is enabled. Also be on the lookout for reverse DNS lookups on the server side. Basically the idea is that they're not expecting an upstream device to manage this traffic. Have a look at this thread:




    Be sure and read the whole thread, it's like an action movie - just missing a car chase or two!



    100% definitely take some captures on the back end server side though. I'd bet that you'll see some sort of off-box lookup going on. Every single time I've heard of multi-second delays like this in a Windows environment it's because of something like this. If you don't mind, would you please post back your findings? It'll help the community immensely.