Re: Kernel Support for Dual Core x86-64?

Robert M. Hyatt, Ph.D.          Computer and Information Sciences
[email protected]                   University of Alabama at Birmingham
(205) 934-2213                  136A Campbell Hall
(205) 934-5473 FAX              Birmingham, AL 35294-1170

On Sat, 17 Sep 2005, Peter Fein wrote:

> Hi all-
> 
> My company is looking to build a cluster of Linux machines to run an
> in-house compute-intensive application.  We're strongly considering
> dual-processor, dual-core x86-64 systems for cost reasons (shared
> hardware, less rack space).  I have a few questions, and despite
> extensive googling, haven't been able to find satisfactory answers.
> 
> 1. Can a dual core chip run two os-level *processes* simultaneously or
> only two threads from the same process?  Our processes (written in
> Python, with parts in C) are completely independent of each other.
> 


yes.   A dual-core is two separate processors, on a single NUMA "node".  
You can create two processes any way you want, from threads, to fork() 
to simply typings a&; b&, and the two processes will run on the two cpus 
just fine.  I've run on a quad 875 system at AMD, and the box appears to 
have 8 cpus since each processor is dual-core.  It runs like a bat out 
of Egypt also...


> 2. What sort of speed up relative to a dual processor, single core
> system could I expect to gain?  I realize this is highly application
> dependent and I'm not looking for any hard figures.  I just want to get
> a rough idea to see if the additional cost is worth it.
> 



technically it is 2x faster.  practically it depends on the application.  
My chess program runs at 2x the nodes per second, but the actual speedup 
is around 1.7-1.8X, due to internal search issues that have nothing to 
do with the dual-core.  In fact, If I run on a dual-core I get about 
1.7-1.8X, if I run on a dual-processor single-core opteron box (say 2 
850's or whatever) I get 1.7-1.8X speedup.  So I can barely tell the 
difference.



> 3. Any suggestions on compiler version?  We're currently using GCC-3.3.6
> on our 32 bit boxes.
> 

I use SuSe on opterons as does AMD on the ones they let me test on.  gcc 
3.3 is plenty good enough...





> Thanks in advance!
> 
> -- 
> Peter Fein                 [email protected]                 773-575-0694
> 
> Basically, if you're not a utopianist, you're a schmuck. -J. Feldman
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-smp" in
> the body of a message to [email protected]
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>