Forum Discussion

komei_100613's avatar
komei_100613
Icon for Nimbostratus rankNimbostratus
Oct 30, 2012

Growing TMM memory usage

Hi all,

 

 

Please give me a advice.I find that memory usage of BigIP in our env is growing.

 

I'm afraid that this trend may lead to swapping out / service down.

 

 

Basic info is as followed :

 

 

- BigIP version 9.3.1 Build 74.0 / BIgIP-3400

 

 

- (liner) memory growing was started in this 3weeks ago.

 

 

- this BigIP box is serving for HTTP/HTTPS access from the

 

　 Internet. Server Nodes under Virtual Servers have very low

 

　 loads.

 

 

- TMM memory usage is now grown to x2.

 

Host Mem Usage is stable.

 

 

- NewConnections/sec, ActiveConnections, Throuput have no change

 

from memory usage growing started.

 

 

- In detail of active connection graph on Big-IP web interface,

 

"server connection"is growing, on the other hand,

 

"client connection" is stable.

 

Before this memory usage growing was started "server connection"

 

and "client connection" were same numbers.

 

Now, "server connection" is twice as bigger as "client connection"

 

 

I doubht the growing memory usage is caused by diffs in num of

 

"server connection" and "client connection" and growing "server

 

connection".

 

 

Any advice or information is appreciated, thanks.

 

 

basic
getting started
new to f5

3 Replies

  • nitass's avatar
    nitass
    Icon for Employee rankEmployee
    Oct 30, 2012
    just curious how come server-side active connection is higher than client-side.

     

     

    if you have qkview which was taken when memory usage was normal, you may compare bigpipe memory show output to see what different is.

     

     

    anyway, i suggest opening a support case, so support engineer can assist you. there are a number of memory leak issue in the past.

     

     

    e.g.

     

    sol10995: TCP connections that do not terminate properly may become zombie connections and cause TMM to leak memory

     

    http://support.f5.com/kb/en-us/solutions/public/10000/900/sol10995.html
  • What_Lies_Bene1's avatar
    What_Lies_Bene1
    Icon for Cirrostratus rankCirrostratus
    Oct 30, 2012
    It might also be worth checking you don't have any server-side profiles configured with an indefinite timeout.
  • komei_100613's avatar
    komei_100613
    Icon for Nimbostratus rankNimbostratus
    Oct 30, 2012
    Thank you for your advices.

    checking you don't have any server-side profiles configured with an indefinite timeout.

    we are using default idletimeout(300sec)

    sol10995: TCP connections that do not terminate properly may become zombie connections and cause TMM to leak memory

    I checked num of close-wait sessions. 

    
 bigpipe tcp show
GLOBAL TCP STATISTICS --
    (open, close-wait, fin-wait, time-wait) = (8406, 160456, 250, 732)
 :

    mmmm... CLOSE-WAIT session seems too many.

    Most of our clients are Android devices(smart phones). They may be end up in bad tcp close procedure,

    because of weak wave in tunnel, underground,etc...but num of server connection is bigger than client

    connection... this point is still unclear and unreasonable for me.

    If you know which connection is in CLOSE-WAIT, please tell me.

    I would like to know IP address of the peer, will clearify server session or client session.

    Anyway, I will also ask to support person too.

    Thanks.