Re: SMP: process-specific options

On 02/23/2010 02:01 AM, Kinkie wrote:
> On Tue, Feb 23, 2010 at 4:54 AM, Alex Rousskov
> <rousskov_at_measurement-factory.com> wrote:
>> On 02/21/2010 10:36 PM, Amos Jeffries wrote:
>>
>>> Is there any kind of kernel core labeling? that would be nicer to link
>>> into than a blind core=N if it exists.
>>> /just a thought.
>> I am not an expert on this, but judging by man sched_setaffinity(2) and
>> man getcpu(2), processors (i.e., cores) are identified by "unique small
>> integers". It is possible that those integers are not sequential on some
>> advanced systems, I guess, but I am not sure we should worry about it
>> now. Any other opinions?
>
> We could let the kernel scheduler decide what's best (aka ignore the issue) :)

Yes, that would be the default. However, I am certain that the default
will not be enough in many high-load deployments, which is the focus of
SMP work.

The scheduler helps with CPU affinity. However CPU affinity
configuration is only a small part of the problem. We need process
naming/numbering/identification for many other, more important reasons
such as having processes dedicated to certain tasks or configuration
subsets.

Thank you,

Alex.
Received on Tue Feb 23 2010 - 15:22:16 MST