[GE users] Job time slowed with Hyperthreading ?

dagru d.gruber at sun.com
Tue Feb 2 10:34:30 GMT 2010

If you use SGE 6.2u5 the correct amount of cores and sockets
per host are reported (see qhost -cb or qstat -F m_core). When you
are using the core binding switch -binding you can let the jobs run
on complete cores (in case of hyperthreading or CMT it will be bound on all
processor numbers representing the selected cores) exclusively.
Isolating jobs from another in this way has also other advantages
like preventing cache pollution and memory access benefits.

But in any case, how many processes can actually run on the host depends
on your slot configuration and not on the amount of reported CPUs.
You may reduce the amount of slots to 4.



On 02/02/10 11:12, umanga wrote:
> Greetings ,
> Some of my execution hosts have i7 processors (4 cores with 
> HyperThreading ) and SGE show that each of this machine has 8 CPUs.
> I noticed that similar jobs take much more time to finish on these 
> processors than on a processors without HT.
> So I disabled HT in CMOS ,but still SGE shows 8 CPUs for this hosts.
> Is there any performance issue with Hyperthreading?
> How can I enable actual physical cores (4) for these hosts?
> Thanks in advance.
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=242586
> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].


To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].

More information about the gridengine-users mailing list