Re: CPU Time slices, technotes.

Remember that on multi-processor Sun boxes vm/mpstat over calculate a variety of the numbers and the more processors you have the worse it gets. Also you can not get round the problem by querying the kernel directly - the problem is the kernel values are not maintained correctly. Solaris 7 has gone a long way to resolve it but AFAIK it still isn't fixed, supposedly Solaris 8 alpha has got round it On the monitoring tools the best IMHO by far is the SEToolkit >Art: > >How are you measuring your results? What tools are you using to do this - >sar, mpstat, vmstat, proctool? > >TIA > >Steve Romankiw "Art S. Kagel" wrote: > David Tamburin wrote: > > > > This is a multi-part message in MIME format. > > > A few days ago I saw a message by Art S. Kagel. It mentioned changing > > some kernal parameters to change timeslice to improve performance. > > Could anyone tell me where I can dig this information up? > > OK Here are the Kernel parameters to adjust the Solaris system clock > and timeslice down form the default 1/10 sec to 1ms. In /etc/system: > > set hires_tick=1 # or 2 - Number of system ticks per timeslice- 1 ms > set hires_hz=1000 # Number of system ticks per second, ie set to ms > > Without this change you will be hard pressed to push the machine > beyond about 65% utilization. After the change you can add processes > and push utilization above 92%! > > Art S. Kagel