ltm unit running at 100% after upgrade to 11.5.3

some output : Tasks: 401 total, 20 running, 381 sleeping, 0 stopped, 0 zombie Cpu(s): 68.7%us, 26.2%sy, 1.3%ni, 3.0%id, 0.2%wa, 0.1%hi, 0.6%si, 0.0%st Mem: 8189868k total, 8092768k used, 97100k free, 103412k buffers Swap: 1048504k total, 262332k used, 786172k free, 447220k cached

 

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND

 

18691 root RT 0 6327m 151m 133m S 30.9 1.9 1770:09 tmm

 

18690 root RT 0 6327m 151m 133m S 28.3 1.9 1855:55 tmm

 

7595 root 20 0 183m 83m 15m R 19.8 1.0 873:24.36 mcpd

 

8889 root 20 0 40712 22m 12m R 11.4 0.3 879:25.49 bigd

 

7113 root 25 5 94348 38m 35m R 5.2 0.5 284:34.64 merged

 

21036 root 20 0 151m 75m 24m S 3.9 0.9 11:53.55 iControlPortal.

 

typically both tmm processes are taking most cpu. (it's 69OO platform we are running on) This is behavior we see continuously. On 2nd unit tmm is consuming less cpu & other processes like mcpd and bigd can sometimes take more cpu than tmm. On first unit this is never the case , both tmm process always are taking most cpu.

 

We are also working with traffic groups , so traffic is crossing on both units units in HA-group. 2nd units has also a raised cpu , but not hitting critical values. 1st unit is more impacted & is slowing down significantly. Main difference between 1st & 2nd unit is that , on 1st unit we are running ltm setups with virtual servers & pool members in directly connected subnets. On 2nd unit most setup have pool members in routed subnets, so not directly connected. (we are using snat for these setups) It's 1st unit who is mainly suffering from increased cpu .

 

We already started cleaning in order to reduce UP/down events in log-files, but this has no impact. On advise of F5 we also adapted some idle timeout timers in snat translation addresses , but this didn't have any impact.