- Tested on:
- Linux 2.4
- RedHat 8 [DON'T USE IT! see below]
- RedHat 9
- RedHat Enterprise 3
- Linux 2.6
- Fedora Core 2
- One some versions of RedHat 8 with glibc??? there is a fatal bug
in the pthread library. Don't use RH8!
- RedHat 9 and RedHat Enterprise 3 seem to have unusually high latency for nanosleep()
as compared to Fedora Core 2. Perhaps this is a 2.4 versus 2.6 kernel difference.
This can adversely affect applications like mcctx. You can mitigate the effects by
using the -sched option to command line programs , for example: '-sched RR 1'.
If you are writing your own C programs you can set the aw_monitor_sched_t structure
in the monitor params to values supported by sched_setscheduler().
See the man page for sched_setscheduler.
- The signal latency in the 2.6 kernels is MUCH lower than the 2.4 kernels.
If you are using the timing stats (e.g., round trip times for ping) in some of the handlers
and want the most accurate results, use a 2.6 kernel. Similarly, high performance
network applications will do much better with a 2.6 kernel.