Re: Which dual opteron?

From: Anton Ertl (anton_at_mips.complang.tuwien.ac.at)
Date: 10/08/04 

Date: Fri, 08 Oct 2004 20:55:56 GMT

Brian Hall <brihall@nowhere.org> writes:
>On 2004-10-08, Anton Ertl wrote:
>> Chris Cox <ccox_nopenotthis@airmail.net> writes:
>>>Al Dykes wrote:
>>>...
>>>>
>>>> All opeterons are numa architecture.
>>>
>>>Well... the cheaper dual boards use a shared bus to a common
>>>bank of memory so in the spirit of numa.. I'd say I don't
>>>think so. The higher end boards are the ones with the
>>>separate banks for each processor. Perhaps the shared
>>>bus is still numa somehow??
>>
>> Certainly a single process can see different (i.e., non-uniform)
>> memory access times on such machines, although now at different times
>> (if the scheduler changes the CPU of the process) rather than for
>> different memory areas.
>>
>> The consequences for the software are simpler than for normal NUMAs
>> (the scheduler should prefer one CPU when only one process is ready),
>> but normal NUMA optimizations (allocate memory close to the process
>> and schedule the process close to its memory) should also degenerate
>> to this simpler policy on such a machine.
>
>So on a board that does this (like my MSI K8T Master2 FAR), should I
>enable NUMA or not in the kernel config? If there is no advantage to
>doing so, I would think it might be better to leave it disabled, since
>the NUMA code path is less used/tested.

I would expect a slight speedup from enabling NUMA optimizations in
some situations (load<2), but I am _not_ speaking from experience, I
don't know if these optimizations will do what I expect, or what
effect they have on the stability.

- anton 

-- 
M. Anton Ertl                    Some things have to be seen to be believed
anton@mips.complang.tuwien.ac.at Most things have to be believed to be seen
http://www.complang.tuwien.ac.at/anton/home.html

Relevant Pages

  • Re: forget coorporative multitasking in .NET ?
    ... However than came IBM with "multitasking", probably with the same reasons as ... with multitasking you had to tell what part of memory would ... > the processor gave execution back to the scheduler. ... > physically halted as another thread was given the CPU. ...
    (microsoft.public.dotnet.languages.csharp)
  • Re: forget coorporative multitasking in .NET ?
    ... > This makes me curious, I know the term multitasking as an IBM term, what was ... > multitasking you had to tell what part of memory would be used and therefore ... the processor gave execution back to the scheduler. ... time, after that, the scheduler simply took teh CPU away from the thread ...
    (microsoft.public.dotnet.languages.csharp)
  • Re: [Lse-tech] [PATCH] cpusets - big numa cpu and memory placement
    ... > Ooooh, CPU 3 over there looks heavily loaded, I'll steal something. ... memory allocated that hurt on our big boxes - such as blowing out one ... The kernel allocator and numa placement policies don't really integrate ... have scheduler and allocation policies handled on a "per-domain" basis, ...
    (Linux-Kernel)
  • [patch] scheduler fix for 1cpu/node case
    ... after talking to several people at OLS about the current NUMA ... Fact is that the current separation of local and global balancing, ... Especially in the corner case of one CPU per node this is ... scheduler while keeping NUMA memory management on. ...
    (Linux-Kernel)
  • Re: Multicore Is Bad News For Supercomputers
    ... NUMA designs from the early wildfire days, but if the OS and App are ... just portions of memory attached along with CPUs at nodes in the interconnect fabric; if the memory address a CPU references does not correspond to local memory, the memory reference goes across the interconnect between nodes via high-speed router hardware at each node and is accessed remotely. ...
    (comp.os.vms)