[R-sig-hpc] parallel and openblas
simon.urbanek at r-project.org
Wed Apr 25 20:38:53 CEST 2012
FWIW: Based on this discussion, I have added the capability to manage CPU affinity mask in R-devel. In the short term you can use mcaffinity() to control the affinity yourself, so for example mcaffinity(1:ncores) will allow R to run on any core or you can restrict your parallel jobs to certain CPU ranges. Eventually, R/parallel should handle that for you by partitioning cores to child processes (we are missing mandatory core detection and tracking to do that), for now the closest you get is mc.affiinity in mcparallel() to spread children yourself.
On Apr 25, 2012, at 12:54 PM, Claudia Beleites wrote:
> I agree with Simon.
> - Not restricting the children doesn't lead to problems with both
> implicit and explicit parallelization for me:
> E.g. if I decide to run explicitly 3 parallel threads and have 12 cores,
> I use GOTO_NUM_THREADS 4, and that works nicely.
> - if users do not have access to taskset, then restriction to 1 core is
> certainly more problematic than no restriction.
> Am 25.04.2012 17:21, schrieb Simon Urbanek:
>> On Apr 25, 2012, at 11:10 AM, Dirk Eddelbuettel wrote:
>>> Sorry to follow-up on my own post from minutes ago, but maybe it is better
>>> the way it is as it allows to explicitly spread tasks each of which would
>>> limit itself to one core. Isn't this the best way to avoid "clogging" when
>>> we use, say, snow to get to eight cores, and each of the eight cores wants to
>>> do BLAS work on eight cores... That was after all the initial issue in
>>> Martin's email.
>>> And when I use the affinity trick (now breaking the single line for display)
>>> edd at max:~$ r -e 'library(parallel); \
>>> system(sprintf("taskset -p 0xffffffff %d", Sys.getpid())); \
>>> cores <- detectCores(); \
>>> print(cores); \
>>> mclapply(1:cores, function(i) repeat sqrt(3.14159), mc.cores=cores)'
>>> pid 24436's current affinity mask: 1
>>> pid 24436's new affinity mask: ff
>>>  8
>>> all is well -- eight cores humming along just fine per htop under OpenBLAS.
>>> Maybe this is actually better as it effectively gives us a run-time toggle?
>> I'd argue that the problem is affinity of 1 by default - that is certainly not what you want or expect and bad in most cases. It would make much more sense to restrict the *children* to one CPU each rather than the parent process so that forked processes stay on their cores - that is IMHO the only place where setting affinity makes any sense at all. Also the system(..) trick is ugly and highly system-specific (you don't even know if you can access taskset). I would consider OpenBLAS' affinity setting as a pretty bad bug (certainly form user's point of view).
>> R-sig-hpc mailing list
>> R-sig-hpc at r-project.org
> Claudia Beleites
> Institute of Photonic Technology
> Albert-Einstein-Str. 9
> 07745 Jena
> email: claudia.beleites at ipht-jena.de
> phone: +49 3641 206-133
> fax: +49 2641 206-399
> R-sig-hpc mailing list
> R-sig-hpc at r-project.org
More information about the R-sig-hpc