[R-sig-hpc] Difference between PSOCK and MPI?
matloff at cs.ucdavis.edu
Sat Apr 13 19:39:25 CEST 2013
A number of problems can arise for the unwary. Having more than one MPI
implementation installed can create path problems, for instance. Some
implementations take over all the cores even when idle. One must
arrange passwordless logins. Error messages are often subliminal.
As I said, none of this is a (big) problem for power users, but I
believe there are lots of people who need parallel computation in R but
are not power users.
On Sat, Apr 13, 2013 at 09:57:08AM -0500, Dirk Eddelbuettel wrote:
> On 11 April 2013 at 13:56, Norm Matloff wrote:
> | I think Simon has a good point here. MPI is great, but it can be very
> | tricky to set up and interface to R.
> | Other than the real power users, who won't use snow/parallel for their
> | power usage anyway, I think PSOCK is just fine.
> | Even we fans of parallel programming don't necessarily want to squeeze
> | every last ounce of speed from a program, at major expense of
> | convenience. Note that even with MPI, there are various versions and
> | configurations to tweak if one really wants to do so; my guess is that
> | most people don't do so.
> With some Linux distributions it can be pretty easy.
More information about the R-sig-hpc