Re: svn commit: r594726 - in /spamassassin/trunk: build/nightlymc/clienthosts masses/rule-qa/nightly-slave-start

[Justin Mason <jm...@jmason.org>]

yep, /proc/cpuinfo says 16.  Should I reduce that back down to 8?
(I haven't had much experience with HT yet)

--j.

Kevin A. McGrail writes:
> Answering only for talon1, yes.  It is a quad processor box with dual-core 
> 3GHz Xeon CPUs.  Thought it may come up as 16 cores because of 
> HyperThreading.
> 
> regards,
> KAM
> 
> >> +jm@talon1.pccc.com:-j8
> >> +jm@infiltrator.stdlib.net:-j8
> >
> > Are those both 8 core machines?  Non-net checks run the fastest with a 
> > single job per core.

Re: svn commit: r594726 - in /spamassassin/trunk: build/nightlymc/clienthosts masses/rule-qa/nightly-slave-start

[Loren Wilton <lw...@earthlink.net>]

Having far more experience than I need on multiproc systems, the answer is 
"it depends".  In all probability having the extra 8 threads running will 
result in some processor speed increase.  It will be less than double, so 
1.0 < x < 2.0.

Of course if this spawns N more instances of SA, you will probably run out 
of memory and thrash.

        Loren

----- Original Message ----- 
From: "Kevin A. McGrail" <ke...@thoughtworthy.com>
To: "Justin Mason" <jm...@jmason.org>
Cc: "Daryl C. W. O'Shea" <sp...@dostech.ca>; "SpamAssassin Dev" 
<de...@spamassassin.apache.org>; "Justin Mason" <jm...@jmason.org>
Sent: Wednesday, November 14, 2007 7:45 AM
Subject: Re: svn commit: r594726 - in /spamassassin/trunk: 
build/nightlymc/clienthosts masses/rule-qa/nightly-slave-start


> Technically or theoretically, telling it you have 16 and sending it more 
> information than CPUs should allow the processor to handle the threading 
> in ways that will give a speed increase.
>
> Unfortunately, the engineer in me is laughing (cackling, actually) and 
> wondering if a speed test of the command at 16 and the command at 8 using 
> even something simple like time would be in order.
>
> Regards,
> KAM
>
>
>> yep, /proc/cpuinfo says 16.  Should I reduce that back down to 8?
>> (I haven't had much experience with HT yet)
>>
>> --j.
>>
>> Kevin A. McGrail writes:
>>> Answering only for talon1, yes.  It is a quad processor box with 
>>> dual-core
>>> 3GHz Xeon CPUs.  Thought it may come up as 16 cores because of
>>> HyperThreading.
>>>
>>> regards,
>>> KAM
>>>
>>> >> +jm@talon1.pccc.com:-j8
>>> >> +jm@infiltrator.stdlib.net:-j8
>>> >
>>> > Are those both 8 core machines?  Non-net checks run the fastest with a
>>> > single job per core.
>>
> 

Re: svn commit: r594726 - in /spamassassin/trunk: build/nightlymc/clienthosts masses/rule-qa/nightly-slave-start

["Kevin A. McGrail" <ke...@thoughtworthy.com>]

Technically or theoretically, telling it you have 16 and sending it more 
information than CPUs should allow the processor to handle the threading in 
ways that will give a speed increase.

Unfortunately, the engineer in me is laughing (cackling, actually) and 
wondering if a speed test of the command at 16 and the command at 8 using 
even something simple like time would be in order.

Regards,
KAM


> yep, /proc/cpuinfo says 16.  Should I reduce that back down to 8?
> (I haven't had much experience with HT yet)
>
> --j.
>
> Kevin A. McGrail writes:
>> Answering only for talon1, yes.  It is a quad processor box with 
>> dual-core
>> 3GHz Xeon CPUs.  Thought it may come up as 16 cores because of
>> HyperThreading.
>>
>> regards,
>> KAM
>>
>> >> +jm@talon1.pccc.com:-j8
>> >> +jm@infiltrator.stdlib.net:-j8
>> >
>> > Are those both 8 core machines?  Non-net checks run the fastest with a
>> > single job per core.
>