vxfsd and threads

[KPS]

We've been investigating some issues with an Application we have running on our rx8640 and it's inability to initiate new processes/jobs now and again. We're running HP-UX 11.23. With the help of kcalarm and kcusage, we've been closely watching kernel parameter usage and we have come to find out that the only tunable that looks like it might be getting close to the threshold is max_thread_proc. Our value is set to the default of 256. After a little more diagnosis, we feel that vxfsd is more than likely the culprit and it's getting dangerously close to the max_thread_proc value limit we have set. I realize I can bump this higher and this is a dynamic kernel tunable.

My question is if there's any kind of consensus around vxfsd and it using 231, 246, 247 threads per proces and if this might be normal or not? This may be too abstract a question to ask, but I just wanted to see if some folks could comment based upon any experiences they've had? Additionally if anyone would have any insight as to what might be making vxfsd behave this way with as many threads per process that it's using, I'd be interested in that too.

Thanks in advance,
/KPS

[Emil Velez]

how many filesystems does the system have ?

have you changed any of the default vxtunefs parameters regarding streams ?

This may increase the number of vxfsd threads.

[KPS]

64 mounted filesystems.

We do have some option specific items on a 3 filesystems in the /etc/vx/tunefstab and they are as follows:

discovered_direct_iosz=512K

In addition to that we have our journaling filesystems (3 filesystems) setup with the following mount options if this might make any difference:
rw,suid,mincache=direct,largefiles,delaylog,datainlog 0 2

Thanks,
KPS

[likid0]

Hi KPS,

We normally pump up max_thread_proc to 512, on all servers just to keep clean of touching limits.

I just checked on some of our servers that have quite a lot of IO, and vxfsd has lots of threads runnig, on some cases between 250-400.

I'll check out the hours I get the peaks, and see whats going on.

[KPS]

Thanks Daniel, that's good information. We're just about to bump up our default from 256 to 512. I think the info you gave me is just some extra reassurance on ideas of increasing this.

Many thanks, i think we're all set...

/KPS