On Wed, Mar 20, 2013 at 2:26 PM, Matt <[email protected]> wrote:
> On Wed, Mar 20, 2013 at 8:34 AM, Ben Noordhuis <[email protected]> wrote:
>>
>> What numbers are you seeing?  large_number / number_of_workers is what
>> you should expect to see: a fairly constant total throughput with
>> per-worker throughput inversely proportional to the number of workers.
>
>
> In case my description wasn't clear, that's NOT what I'm seeing.
>
> I'm only logging the send rate. I don't care what the workers receive. I
> would expect the send rate to be constant since the master is always doing
> the same amount of work.
>
> Here's some numbers:
>
> Rate for 1 workers: 249353
> Rate for 2 workers: 220569
> Rate for 3 workers: 213680
> Rate for 4 workers: 205965
> Rate for 8 workers: 147463
>
> (the machine has 12 cores)
>
>>
>> Re: hyperthreading, if that's enabled test with num_workers = num_cpus
>> / 2 (and pin each worker to a separate physical CPU.  Recent kernels
>> do a pretty good job of managing that themselves, though.)
>
>
> As I said in the original post, I did that to eliminate hyperthreading being
> a slowdown.

You don't.  If the machine has 12 HT CPUs, you should test with (12 /
2) == 6 processes.

(I said that wrong in my previous email: processes, not workers.  So
in this case 5 workers plus the master.)

Oh, and disable Turbo Boost(TM) - it tends to introduce a lot of
variance in multi-core benchmarks.

-- 
-- 
Job Board: http://jobs.nodejs.org/
Posting guidelines: 
https://github.com/joyent/node/wiki/Mailing-List-Posting-Guidelines
You received this message because you are subscribed to the Google
Groups "nodejs" group.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to
[email protected]
For more options, visit this group at
http://groups.google.com/group/nodejs?hl=en?hl=en

--- 
You received this message because you are subscribed to the Google Groups 
"nodejs" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/groups/opt_out.


Reply via email to