Forum Discussion

Deon's avatar
Deon
Icon for Nimbostratus rankNimbostratus
Mar 16, 2011

What Happens If 6400 Pushed Beyond Its Capacity?

Hi All. I have a 6400 HA pair running 10.2.0 LTM and very recently added ASM with seven ASM policies. Total CPU use varies between ~65% and 100% after enabling ASM. Several of our websites that have ASM policies will experience a 20x increase in traffic later this year. ASM was not part of the picture last season when we experienced the expected spike in traffic to these websites. I am not certain if the 6400 will be able to handle the load with ASM turned on now. What occurs when a dual CPU system running LTM and ASM runs out of CPU? I presume LTM side of things will be alright because my understanding is that TMM grabs one whole CPU when it starts up. I would expect CPU resources for ASM to be the issue point. What is the behaviour one would expect to see with an ASM enabled website in a condition where the BigIP is pushed to 100% CPU use? Does traffic slow down because of ASM processing or does ASM drop some traffic/connections or is ASM possibly rendered ineffective but LTM traffic is not affected?

 

 

As you might imagine, I am working on building the case for replacing these 6400s with new equipment. I am trying to establish the urgency and what-if scenarios of deciding to leave it as-is vs replacement. Thanks!

 

  • Deon,

     

     

    In v10.0+, the TMM does not run on just one CPU, it grabs both just like it does for the other platforms. This means that TMM can consume up to 90% of both CPUs rather than 100% of one. This results in a nice performance boost in most regards (see SOL9763: http://support.f5.com/kb/en-us/solutions/public/9000/700/sol9763.html?sr=13375974). However it also means that tasks that were previously relegated to the second CPU will receive much less CPU time than before.

     

     

    As for what the effect will be, I expect traffic will slow radically, probably to the point of some browsers timing out (and certainly some users). If CPU is actually maxed out you will also see TMM unable to respond to connection requests. It won't result in connection rejections by default, but connections that don't use ASM will timeout as well. Further, health checks will start to fail because the health check daemon, bigd, doesn't get enough CPU time to send health checks or process the replies. That will drastically decrease user experience because even connections that weren't timing out will be interrupted as the node they are connected to is marked down. Limiting TMM to only 90% helps this situation a lot, but under extreme pressure it will still happen.

     

     

    Hope this helps.

     

     

     

    --jesse
  • Deon's avatar
    Deon
    Icon for Nimbostratus rankNimbostratus
    Thanks much jesse for your reply. Very helpful information and this better understanding will aid in our decision to upgrade these devices.

     

     

    -Deon