I've only tried multicore. See here:

http://news.gmane.org/gmane.comp.lang.r.datatable/cutoff=326

If data tables are only used on the inside of the "for loop", it
should be pretty easy. If you need data tables grouping operations to
be run in parallel, it's quite a bit harder.

- Tom

On Fri, Dec 31, 2010 at 9:22 AM, Matthew Dowle <[email protected]> wrote:
> Tom might be your man on that.
> What I'd say is that if the 7.5sec wasn't ok for one grouping then we
> might be a bit stuck, but in your case it should be ok using a parallel
> wrapper through the 500 groupings. I'd need a toy example to go further
> with this one please to see why there's no apparent speed up.
> Matthew
>
>
> On Wed, 2010-12-29 at 11:25 -0600, Damian Betebenner wrote:
>> All,
>>
>>
>>
>> Does anyone have experience using data.table in parallel using
>> doMC/foreach?
>>
>>
>>
>> I have a data.table with approximately 3.5 million rows and am
>> calculating different summaries (e.g., medians and counts) on some of
>> the variables across approximately
>>
>> 500 distinct groupings (same j variable with 500 different by groups).
>> My thought was to run the different analyses in parallel on my
>> multi-core machine and get a good performance
>>
>> boost but thus far it doesn’t work any faster.
>>
>>
>>
>> The data.table performance is great for each by grouping (about 7.5
>> seconds), but doing this 500 times take a while.
>>
>>
>>
>> Anyone have experience along these lines.
>>
>>
>>
>> Best,
>>
>>
>>
>> Damian
>>
>>
>>
>>
>>
>>
>>
>>
>> _______________________________________________
>> datatable-help mailing list
>> [email protected]
>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/datatable-help
>
>
> _______________________________________________
> datatable-help mailing list
> [email protected]
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/datatable-help
>
_______________________________________________
datatable-help mailing list
[email protected]
https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/datatable-help

Reply via email to