Huge CPU usage on multithread processes

[inactive account]

My applications use a very large amount of cpu on HPUX.
THey consist on different multi-threads processes communicating with each other en sockets.
When trying to run a benchmark I take all CPU. The same application on LINUX Pentium4 box takes 5 % of the CPU.
I think that maybe the loss is on the way HPUX deals with POSIS PTHREAD implementation.
Does anybody have informations about tuning the threads on HPUX ?

Thanx to anyboby that can help!

[harry d brown jr]

Start here:
http://forums.itrc.hp.com/cm/components/FileAttachment/0,,0x33c03fa720f3d5118ff40090279cd0f9,00.pdf

live free or die
harry

[harry d brown jr]

And

http://h21007.www2.hp.com/dspp/tech/tech_TechDocumentDetailPage_IDX/1,1701,984,00.html

http://h21007.www2.hp.com/dspp/tech/tech_TechDocumentDetailPage_IDX/1,1701,970,00.html

http://h21007.www2.hp.com/dspp/tech/tech_TechSingleTipDetailPage_IDX/1,2366,158,00.html

http://www.hp.com/hpbooks/prentice/ptr_0131900676.html

live free or die
harry

[John Bolene]

You need to tell us more about your application and the machines that you ran these tests on.

The clock speed on HP Unix machines is slower than runs on PC's, but HP gets more performance out of the same clock cycle.

I will have to relate a few cpu tests I have run myself. These involve how long it takes to run a seti chunk on each cpu. Seti processing is 99.9% cpu related and only uses one processor since it is not multi-threaded.

200 Mhz MMX - 40 hours
120 Mhz C110 - 30 hours
500 Mhz Pentium - 11.5 hours
800 Mhz celeron - 12 hours
1.3G AMD - 7.5 hours
1.4G AMD - 7 hours
875 Mhz superdome - 4.5 hours
2400 AMD - 3 hours
2700 AMD - 2.6 hours

So if you are running Linux on a PC with a processor > 2.1G, you will get faster run time than on a superdome. The superdome has the edge in the number of processors it can run and the amount of I/O it can connect to.