Thanks Reuti, I didn't know about the MPI_REMSH env var! And by "known unsupported", we mean the MPI implementation does not support tight integration. But if the MPI_REMSH mechanism fixes that, then may be I let others know - some people complained before that Platform MPI sometimes leaves runaway processes running on the nodes.
Rayson On Sat, May 12, 2012 at 11:42 AM, Reuti <re...@staff.uni-marburg.de> wrote: > Am 10.05.2012 um 18:13 schrieb Rayson Ho: > >> Typo, it should be: >> >> The know "UNSUPPORTED" MPI implementations are HP-MPI & Platform MPI > > Why is it declared "unsupported"? We use it for some while with the Quantum > Chemistry packages ADF and TurboMole and it's working out-of-the box. But as > I don't have a native HP-MPI around, I don't know what scripts were added by > the particular vendors only. > > According to the original HP-MPI documentation, with the latest versions of > HP-MPI no longer any special "appfile" or whatever it was called is > necessary, but a list of hosts in the MPICH1 style is sufficient. The startup > mechanism can by routed to `qrsh -inherit ...` by setting "export > MPI_REMSH=rsh" und use the rsh-wrapper like for MPICH1 to use the "builtin" > startup method. > > NB: Platform Computing was acquired by IBM > http://www-03.ibm.com/systems/technicalcomputing/platform.html > > -- Reuti > > >> (ie. Scali + HP MPI - Platform Computing bought the IP from Scali & HP >> a few years ago). >> >> Rayson >> >> >> >> On Thu, May 10, 2012 at 12:08 PM, Rayson Ho <ray...@scalablelogic.com> wrote: >>> Since Hung-Sheng has already answered what MPI implementations are >>> supported, I will try to answer the known unsupported MPI >>> implementations. >>> >>> Also, let me clarify that our definition of "unsupported" is that the >>> MPI implementation does not tightly integrate with Grid Engine - and >>> thus Grid Engine could not kill, checkpoint, perform core binding, and >>> record job resource usage. If an MPI application is able to run >>> outside of Grid Engine, then it should also work as a Grid Engine >>> batch job even without tight integration. >>> >>> The know "supported" MPI implementations are HP-MPI & Platform MPI >>> (ie. Scali + HP MPI - Platform Computing bought the IP from Scali & HP >>> a few years ago). >>> >>> See also: http://blogs.cisco.com/performance/platform_acquires_hp-mpi/ >>> >>> Rayson >>> >>> >>> >>> >>> >>> On Thu, May 10, 2012 at 9:51 AM, Hung-Sheng Tsao Ph.D. >>> <laot...@gmail.com> wrote: >>>> Tight Integration of Parallel Libraries >>>> >>>> Tight Integration of LAM/MPI and SGE >>>> Tight Integration of MPICH and SGE -- With Application Notes >>>> Tight Integration of MPICH2 and SGE >>>> Tight Integration of PVM and SGE >>>> Mvapich (MPICH Infiniband) + Loose/Tight SGE Integration >>>> Sun HPC Cluster Tools parallel jobs (MPI, MPI2, OpenMP) >>>> Tight integration of Open MPI with SGE >>>> >>>> >>>> >>>> On 5/10/2012 9:34 AM, Chi Chan wrote: >>>> >>>> Good point, Walid. As much as so many people hate UNIVA, IMHO, there >>>> can be things learned from them. (In Chinese "三人行,必有我师", and >>>> translation in English it is something like "Everyone and Everything >>>> Around You is a Teacher") >>>> >>>> And since UNIVA said on their website that they are now documenting >>>> the known working configuration of MPI libraries, my question is, what >>>> MPI libraries are known to work with GridEngine, and what is known not >>>> to work with GridEngine? >>>> >>>> --Chi >>>> >>>> On Sun, May 6, 2012 at 5:38 PM, Walid <walid.sha...@gmail.com> wrote: >>>> >>>> The use of postgres could be due to the standby, and replication features, >>>> as well as the use of postgres already by Unisight. I am a Univa Happy >>>> customer so far ;-) >>>> >>>> >>>> On 4 May 2012 03:48, Ron Chen <ron_chen_...@yahoo.com> wrote: >>>> >>>> Univa have "New support for Postgres database job spooling balances speed >>>> of submission with reliability in high volume clusters with lots of small >>>> jobs" as one of the new features of UGE 8.1 on their website. >>>> >>>> We have discussed here before, PostgreSQL spooling was added by Sun, but >>>> was taken out in SGE 6.2 as Sun believes BerkeleyDB (BDB) is good enough. >>>> >>>> If Univa is re-using the code originally developed by Sun, then please >>>> don't take Sun's credit. >>>> >>>> >>>> -Ron >>>> >>>> >>>> _______________________________________________ >>>> users mailing list >>>> users@gridengine.org >>>> https://gridengine.org/mailman/listinfo/users >>>> >>>> _______________________________________________ >>>> users mailing list >>>> users@gridengine.org >>>> https://gridengine.org/mailman/listinfo/users >>>> >>>> >>>> -- >>>> >>>> >>>> _______________________________________________ >>>> users mailing list >>>> users@gridengine.org >>>> https://gridengine.org/mailman/listinfo/users >>>> >> >> _______________________________________________ >> users mailing list >> users@gridengine.org >> https://gridengine.org/mailman/listinfo/users >> > _______________________________________________ users mailing list users@gridengine.org https://gridengine.org/mailman/listinfo/users