[Beowulf] Re: Re: Re: Home beowulf - NIC latencies (Patrick Geoffray)

Ole W. Saastad ole at scali.com
Tue Feb 15 01:28:10 PST 2005

Patrick Geoffray wrote:
> There are more exotic work-arounds, like using 1) and polling at the 
> same time, and hiding the interrupt overhead with some black magic on 
> another processor. The one with the best potential would be to use 
> HyperThreading on Intel chips to have a polling thread burning cycles 
> continuously; it will run in-cache, won't use the FP unit or waste 
> memory cycles. A perfect use for the otherwise useless HT feature. I 
> wonder why nobody went that way...

I have tried this in order to see if we could poll a memory location
for free using Intel HT. I run a kinetics program that is a small
Runge-Kutta stepping of equations simultaneously with a small
loop checking the content of a memory location then issuing a
PAUSE instruction and repeating the loop.

The simple finding is that the kinetics program got somewhat more 
than 70% of the CPU cycles and that the polling waisted close to
30% of the CPU cycles, 30% is not for free.

After this test I finally decided to forever leave HT off.


Ole W. Saastad, Dr.Scient. 
Manager Cluster Expert Center
dir. +47 22 62 89 68
fax. +47 22 62 89 51
mob. +47 93 05 74 87
ole at scali.com

Scali - www.scali.com
High Performance Clustering

More information about the Beowulf mailing list