[GRLUG] FYI - old mysql and new gear
Bob Kline
bob.kline at gmail.com
Thu Sep 22 15:16:47 EDT 2011
On Thu, Sep 22, 2011 at 3:01 PM, Mike Williams <knightperson at zuzax.com>wrote:
> On 9/22/2011 2:31 PM, Michael Mol wrote:
>
>> On Thu, Sep 22, 2011 at 2:29 PM, Mike Williams<knightperson at zuzax.**com<knightperson at zuzax.com>>
>> wrote:
>>
>>> On 9/22/2011 2:20 PM, scott.tanner at comcast.net wrote:
>>>
>>> I'm very familiar with nice, but its the taskset command I was referring
>>> to,
>>> which takes affect at the scheduler:
>>>
>>> http://www.cyberciti.biz/tips/**setting-processor-affinity-**
>>> certain-task-or-process.html<http://www.cyberciti.biz/tips/setting-processor-affinity-certain-task-or-process.html>
>>>
>>>
>>> Regards,
>>> Scott
>>>
>>> That's good stuff to know for the day when I get to play with a box that
>>> has
>>> more than2 actual cores. Regarding turning off hyperthreading, I assume
>>> it's
>>> known which pairs of CPU's are actually the same core? Maybe you could
>>> see
>>> if there's any performance difference running it on 1, 3, 5, and 7 versus
>>> 1,
>>> 2, 3, and 4 (assuming that 1 and 2 are core the first core, 2 and 3 are
>>> the
>>> second, etc).
>>>
>> It's worth noting that the kernel does have support for HT-awareness.
>> I don't know whether Scott's kernel has that configuration option
>> enabled, though.
>>
> Theoretically, yes. In practice, I've heard far more stories of
> hyperthreading making things worse rather than better, but I admit I have no
> first-hand experience with it.
>
>
>
Is hyperthreading more of a hardware
function or a software? The thing that
often kills speedup is branching. If you
can keep a long sequence of computations
going, you get speedup, but that doesn't
always work out in practice either.
What role do compilers have here, if any?
-- Bob
--
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://shinobu.grlug.org/pipermail/grlug/attachments/20110922/475d2e13/attachment.html>
More information about the grlug
mailing list