Forum Discussion
NimbostratusWhat does TMM do when it is CPU bound?
We have a active/standby LTM pair in which BOTH hosts are CPU bound at peak periods. There are 4 cores per pair, the TMM processes are using 99% CPU on both of their two cores on each host. (TMSH 11.6.0)
What does TMM do under these circumstances? Does it drop connections, refuse new connections, or both? Or does it just start paging and get slow like a normal OS?
wsanders_233261Ooops I was getting standby member CPU from the wrong SNMP OID. Look like the standby host is not actually using any CPU, so the CPU on the active host is due to TMM scanning the 20K+ connections in the table.
- wsanders_233261
Just as a warning, we have gone through our git blame for what initiated the pegged CPU, and the root cause was changing to "response-chunking rechunk" from whatever the default is for http profiles, "selective", I think.
This increased our CPU consumption drastically as shown below.
This is a 4 core virtual LTM on a 2150 blade. This blade processes about 1000 reqs/sec with 5 to 10k connections.
So "rechunk" is a very bad idea. The picture shows it all: