[bf1942] 64bit linux BF2 performance stats

James Gurney james at globalmegacorp.org
Wed Jul 20 18:04:05 EDT 2005

On 7/20/2005 3:46 AM, Andreas Fredriksson wrote:
> The only price you pay for CPU migration of a thread is some
> added migration overhead in the kernel which amounts to very little
> time compared to the 10ms (standard) timeslice of most kernels.

Is that still true for NUMA based systems? I'm a little unclear, but my 
understanding was that the main advantage of NUMA was dedicated memory 
per cpu, so migrating from one cpu to the other would lose that 
advantage. Quoting from the (hugely out of date) Linux NUMA page:

"On NUMA systems, optimal performance is obtained by locating processes 
as close to the memory they access as possible.  For most processes, 
optimal performance is obtained by allocating all memory for the process 
from the same node, and dispatching the process on processors on that 
node.  NUMA awareness within the scheduler is necessary in order to 
support locality of processes to memory - primarily by dispatching a 
process on the same node through the duration of the procesess' life."

 From that, I'm concluding that locking at least the most memory 
intensive threads to one cpu would be of benefit on a NUMA system, but 
perhaps I'm misunderstanding the overall picture..


More information about the Bf1942 mailing list