From: Ian Bush on 12 Jul 2010 16:14 On Jul 12, 3:46 pm, n...(a)cam.ac.uk wrote: > In article <6956b18f-159d-469c-948c-eb62fc79b...(a)d16g2000yqb.googlegroups..com>, > Ian Bush <ianbush.throwaway.acco...(a)googlemail.com> wrote: > > > > >Here's the list for DL_POLY_3, CRYSTAL and CASTEP. I don't work on > >GAMESS-UK > >anymore and don't have the code easily to hand, but I can dig it out > >if you are > >interested, the list will be similar to CRYSTAL. VASP is another > >example. I don't > >have the code here, but I would be surprised if it is markedly > >different from CASTEP. > > Thanks very much. Upon updating the table, I realise that I misspoke > earlier - there were in fact FOUR applications that used multiple > communicators - two used groups not MPI_Comm_split. I also seem > to have had bad data for CASTEP, so my remark was a bit off-beam > about the frequency of use for the major applications. Sorry about > that .... > No problem. With regard to CASTEP it depends on how old a version your data refers to. If it is a very old version, yes no more than MPI_COMM_WORLD is used, but more modern versions use quite a deep hierarchy of communicators. This follows my experience - use of communicators is becoming more and more commonplace in MPI codes, mainly to due to the need to exploit hierarchical parallelism since over the last few years the computational node has become relatively much more powerful than the network. For CRYSTAL I can count 5 levels in the hierarchy, and all these levels are currently employed in real calculations. Hence my comments above, Ian > The list won't post, as it is too wide, but please Email me if you > want to see it. > > Regards, > Nick Maclaren.
From: nmm1 on 12 Jul 2010 16:22 In article <473ab577-d7c5-41e8-b61b-1b796a07b2c6(a)y11g2000yqm.googlegroups.com>, Ian Bush <ianbush.throwaway.account(a)googlemail.com> wrote: > >This follows my experience - use of communicators is becoming more >and more commonplace in MPI codes, mainly to due to the need to >exploit hierarchical parallelism since over the last few years the >computational node has become relatively much more powerful than the >network. For CRYSTAL I can count 5 levels in the hierarchy, and all >these levels are currently employed in real calculations. Hence my >comments above, Very interesting! That's useful data, and clearly relevant to the Fortran TR - i.e. a simple one-level team mechanism is NOT enough, and it must be usable hierarchically. I am not going to teach that, because my audience already gets shell-shock, due to being taught all of the core of MPI-1 and some of MPI-2 in three days :-) I do teach MPI_Comm_split, briefly. The same applies to teaching modern Fortran. Covering almost all of the core of Fortran 95 (and bits of 2003) in three days is tough. Regards, Nick Maclaren.
From: Gib Bogle on 12 Jul 2010 21:37
nmm1(a)cam.ac.uk wrote: "...given that it seems to be composed of computer scientists." Now that's not very polite ;-). |