Forum Discussion

Jason_22051's avatar
Jason_22051
Icon for Nimbostratus rankNimbostratus
Jul 16, 2014

"balance" issues with oneconnect

Hi I have a BigIP 3600 running TMOS 9.4.5. My setup is like this:

 

One Virtual server, standard tcp with an HTTP profile (modified only to add a fallback url) A oneconnect profile added to the virtual with the source mask changed to 255.255.255.255. Everything else is default A Universal persistence profile attached to an IRule with the following:

 

when HTTP_REQUEST { if { [HTTP::cookie exists "__appname"] } { persist uie [HTTP::cookie "__appname"] } } when HTTP_RESPONSE { if { [HTTP::cookie exists "__appname"] } { persist add uie [HTTP::cookie "__appname"] } }

 

A pool with 3 servers, using ratio member with each server having a ratio of 1 (round robin essentially)

 

The scenario: We keep 2 of the 3 servers in the pool at all times (the other is forced offline as a hot spare). During a release or issue with one of the other 2 servers, we will bring the third server online. During normal operation, servers A and B get ~5500 connections (because the default on oneconnect is 10K).

 

The problem: When we bring server C in and take server A out, we expect to see the BigIP balance the connections between B and C "evenly" (within an acceptable margin). What we actually see if server B gets 11K+ connections and server C hovers around 50.

 

Anyone know why this is and what, if anything can be done to force the balance. If you need more information, please let me know.

 

TIA

 

application delivery
design
LTM
management
performance
TMOS

2 Replies

Sort By
  • shaggy's avatar
    shaggy
    Icon for Nimbostratus rankNimbostratus
    Jul 16, 2014

    Persistence will almost always result in uneven load distribution among pool members. If user sessions are long-lived, then they will continue to persist to servers A and B even if server C has a much lower connection count. What is the timeout of the universal persistence profile assigned to the virtual server, what is the average/normal user-session duration, and how long does C hover around 50 connections?

     

  • Jason_22051's avatar
    Jason_22051
    Icon for Nimbostratus rankNimbostratus
    Jul 16, 2014

    Shaggy,

     

    Thanks for your response. I realize that persistence will result in uneven distribution. However I don't think the problem here is actually persistence. I think it's oneconnect. Basically the problem as I see it is that oneconnect isn't running against server C. The connections aren't being kept open like they are on A and B