accuracy of timeouts in select/poll in 11.11

[aussel_1]

With HP-UX 11.11, when I use select or poll to suspend a process for several tenths of milliseconds, the accuracy of the wake-up is not good.
The process can be awaken several tenths of milliseconds after the expected value.
The process is run with a "rtprio 100" command.
The computer has a very low CPU use.
If I use usleep instead of select/poll, it works fine.
When I run the same program on HP-UX 11.0, it works fine too.
I apply all the patches concerning select.
What can I do to make the select more precise with HP-UX 11.11 ?

[Steven E. Protter]

Other operating system issues may be in play. Do you have a recently quarterly patch set installed?

If the system is very busy, there may be processes bound to the CPU preventing wake up.

SEP

[Kent Ostby]

There is a known issue in 11.11 where a poll of reused socket increases the wait time by a factor of 10.x.

The fix is in the 11.11 STREAMS patch PHNE_30367 .

[rick jones]

While it should not be "too" much later, keep in mind that those timeouts are of the "guaranteed to to be less than" variety. So, getting say 49 ms on a 40 ms timer would not be surprising.

I suppose you could also try an event port - I'm ass-u-me-ing that your using selelect/poll implies that there are FD's whos read/write status needs to be determined. With the eventport patches installed (/dev/poll) you can do a "man 7 poll" to get the details. Otherwise, select/poll are not really meant to be used as just a sleep mechanism - at least not these days