I have placed the 1.0.6p1 tar ball at ftp://ftp.mcs.anl.gov/pub/mpi/mpich2-1.0.6p1.tar.gz . Jayesh, please create the Windows binaries from this one. Others, please test this version if you have time. We will release it when the Windows binaries are ready. Rajeev
cxx/io/writeshx is failing for sock. -- Pavan On 10/30/2007 03:31 PM, Rajeev Thakur wrote:
I have placed the 1.0.6p1 tar ball at ftp://ftp.mcs.anl.gov/pub/mpi/mpich2-1.0.6p1.tar.gz .
Jayesh, please create the Windows binaries from this one.
Others, please test this version if you have time.
We will release it when the Windows binaries are ready.
Rajeev
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
I ran on shakey and all tests passed. Rajeev
-----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Pavan Balaji Sent: Tuesday, October 30, 2007 4:26 PM To: [email protected] Subject: Re: [mpich2-core] 1.0.6p1 tar ball
cxx/io/writeshx is failing for sock.
-- Pavan
On 10/30/2007 03:31 PM, Rajeev Thakur wrote:
I have placed the 1.0.6p1 tar ball at ftp://ftp.mcs.anl.gov/pub/mpi/mpich2-1.0.6p1.tar.gz .
Jayesh, please create the Windows binaries from this one.
Others, please test this version if you have time.
We will release it when the Windows binaries are ready.
Rajeev
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
I ran it on two nodes. --P On 10/30/2007 04:50 PM, Rajeev Thakur wrote:
I ran on shakey and all tests passed.
Rajeev
-----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Pavan Balaji Sent: Tuesday, October 30, 2007 4:26 PM To: [email protected] Subject: Re: [mpich2-core] 1.0.6p1 tar ball
cxx/io/writeshx is failing for sock.
-- Pavan
On 10/30/2007 03:31 PM, Rajeev Thakur wrote:
I have placed the 1.0.6p1 tar ball at ftp://ftp.mcs.anl.gov/pub/mpi/mpich2-1.0.6p1.tar.gz .
Jayesh, please create the Windows binaries from this one.
Others, please test this version if you have time.
We will release it when the Windows binaries are ready.
Rajeev
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
On Tue, Oct 30, 2007 at 05:35:10PM -0500, Pavan Balaji wrote:
I ran it on two nodes. --P
if the common file system was NFS, release-note this and move on. fcntl locks on NFS often don't work across nodes (do they ever?). ==rob -- Rob Latham Mathematics and Computer Science Division A215 0178 EA2D B059 8CDF Argonne National Lab, IL USA B29D F333 664A 4280 315B
Yes, it is NFS. In that case, should we officially drop support for NFS? -- Pavan On 10/31/2007 10:34 AM, Robert Latham wrote:
On Tue, Oct 30, 2007 at 05:35:10PM -0500, Pavan Balaji wrote:
I ran it on two nodes. --P
if the common file system was NFS, release-note this and move on. fcntl locks on NFS often don't work across nodes (do they ever?).
==rob
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
On Wed, Oct 31, 2007 at 10:36:48AM -0500, Pavan Balaji wrote:
Yes, it is NFS. In that case, should we officially drop support for NFS?
I'd love to, but no. shared file pointers on NFS require group-wide locking. This situation is particularly problematic for NFS, but not impossible. For lots of other MPI-IO workloads (but not all, as we've seen) NFS is adequate. ==rob -- Rob Latham Mathematics and Computer Science Division A215 0178 EA2D B059 8CDF Argonne National Lab, IL USA B29D F333 664A 4280 315B
We have always maintained that NFS should be used for development only, so an MPI-IO implementation that can be used on NFS is still useful, rather than not having one at all. Rajeev
-----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Robert Latham Sent: Wednesday, October 31, 2007 11:35 AM To: [email protected] Subject: Re: [mpich2-core] 1.0.6p1 tar ball
On Wed, Oct 31, 2007 at 10:36:48AM -0500, Pavan Balaji wrote:
Yes, it is NFS. In that case, should we officially drop support for NFS?
I'd love to, but no. shared file pointers on NFS require group-wide locking. This situation is particularly problematic for NFS, but not impossible. For lots of other MPI-IO workloads (but not all, as we've seen) NFS is adequate.
==rob
-- Rob Latham Mathematics and Computer Science Division A215 0178 EA2D B059 8CDF Argonne National Lab, IL USA B29D F333 664A 4280 315B
Yes, it is NFS. In that case, should we officially drop support for NFS?
I'd love to, but no. shared file pointers on NFS require group-wide locking. This situation is particularly problematic for NFS, but not impossible. For lots of other MPI-IO workloads (but not all, as we've seen) NFS is adequate.
Yeah, it works for sequential I/O. It should also work for cases with multiple writers that do not require any locking (e.g., different files). But are there cases in parallel I/O that require locking and ROMIO still works? I'm just trying to narrow down cases that it doesn't work in (or at least might or might not work in). Otherwise, saying "it doesn't work in some cases" might be too vague. -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji
ch3:nemesis passed all mpich2 tests. -d On 10/30/2007 03:31 PM, Rajeev Thakur wrote:
I have placed the 1.0.6p1 tar ball at ftp://ftp.mcs.anl.gov/pub/mpi/mpich2-1.0.6p1.tar.gz .
Jayesh, please create the Windows binaries from this one.
Others, please test this version if you have time.
We will release it when the Windows binaries are ready.
Rajeev
participants (4)
-
Darius Buntinas -
Pavan Balaji -
Rajeev Thakur -
robl＠mcs.anl.gov