MPICH2 1.0.6rc1 thread support
Hi, The README states that and serialized and multiple thread levels are only support by the sock channel, but I can configure the ssm channel to use serialized i.e.: --with-device=ch3:ssm --enable-threads=serialized where --with-device=ch3:ssm --enable-threads=multiple reports a configure error as expected. Is configure correct or the README? BTW some sort of feature matrix for the various devices and channels would be very helpful. Regards James -- ------------------------------------------------------------------------ James S. Perrin, | email: [email protected] Research Computing Services, | web: www.mc.manchester.ac.uk Kilburn Building, The University, | tel: +44 161 275 6945 Manchester, England. M13 9PL. | fax: +44 161 275 0637 ------------------------------------------------------------------------ "The test of intellect is the refusal to belabour the obvious" - Alfred Bester ------------------------------------------------------------------------
Actually serialized and multiple thread levels are supported for sock and nemesis. I have updated the README. For other channels such as ssm, which have no thread support, we currently allow single and funneled. Do you need to use serialized with ssm? Rajeev
-----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of James S Perrin Sent: Friday, August 17, 2007 9:55 AM To: [email protected] Subject: [MPICH2-dev] MPICH2 1.0.6rc1 thread support
Hi, The README states that and serialized and multiple thread levels are only support by the sock channel, but I can configure the ssm channel to use serialized i.e.:
--with-device=ch3:ssm --enable-threads=serialized
where
--with-device=ch3:ssm --enable-threads=multiple
reports a configure error as expected. Is configure correct or the README?
BTW some sort of feature matrix for the various devices and channels would be very helpful.
Regards James -- -------------------------------------------------------------- ---------- James S. Perrin, | email: [email protected] Research Computing Services, | web: www.mc.manchester.ac.uk Kilburn Building, The University, | tel: +44 161 275 6945 Manchester, England. M13 9PL. | fax: +44 161 275 0637 -------------------------------------------------------------- ---------- "The test of intellect is the refusal to belabour the obvious" - Alfred Bester -------------------------------------------------------------- ----------
Hi, We want to use ssm as we sometimes have two specific processes per machine that working as a team. Actaully I think we have been using assuming serialized support at compile time (we're not doing runtime checking of threads_provided, long story don't ask) and it seems to have been working, we hadn't explicitly enabled threading in mpich2 so I gues we've been "lucky" :) Our Master/Slaves app has a threaded Master process that performs serialized MPI calls in two threads. Regards James Rajeev Thakur wrote:
Actually serialized and multiple thread levels are supported for sock and nemesis. I have updated the README. For other channels such as ssm, which have no thread support, we currently allow single and funneled. Do you need to use serialized with ssm?
Rajeev
-----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of James S Perrin Sent: Friday, August 17, 2007 9:55 AM To: [email protected] Subject: [MPICH2-dev] MPICH2 1.0.6rc1 thread support
Hi, The README states that and serialized and multiple thread levels are only support by the sock channel, but I can configure the ssm channel to use serialized i.e.:
--with-device=ch3:ssm --enable-threads=serialized
where
--with-device=ch3:ssm --enable-threads=multiple
reports a configure error as expected. Is configure correct or the README?
BTW some sort of feature matrix for the various devices and channels would be very helpful.
Regards James -- -------------------------------------------------------------- ---------- James S. Perrin, | email: [email protected] Research Computing Services, | web: www.mc.manchester.ac.uk Kilburn Building, The University, | tel: +44 161 275 6945 Manchester, England. M13 9PL. | fax: +44 161 275 0637 -------------------------------------------------------------- ---------- "The test of intellect is the refusal to belabour the obvious" - Alfred Bester -------------------------------------------------------------- ----------
-- ------------------------------------------------------------------------ James S. Perrin, | email: [email protected] Research Computing Services, | web: www.mc.manchester.ac.uk Kilburn Building, The University, | tel: +44 161 275 6945 Manchester, England. M13 9PL. | fax: +44 161 275 0637 ------------------------------------------------------------------------ "The test of intellect is the refusal to belabour the obvious" - Alfred Bester ------------------------------------------------------------------------
James, OK, we will allow thread serialized for ssm and shm channels as well (in 1.0.6). We will also provide a feature matrix. See the example in http://www-unix.mcs.anl.gov/~balaji/support.html . Regards, Rajeev
-----Original Message----- From: James S Perrin [mailto:[email protected]] Sent: Friday, August 17, 2007 12:03 PM To: Rajeev Thakur Cc: [email protected] Subject: Re: [MPICH2-dev] MPICH2 1.0.6rc1 thread support
Hi,
We want to use ssm as we sometimes have two specific processes per machine that working as a team.
Actaully I think we have been using assuming serialized support at compile time (we're not doing runtime checking of threads_provided, long story don't ask) and it seems to have been working, we hadn't explicitly enabled threading in mpich2 so I gues we've been "lucky" :) Our Master/Slaves app has a threaded Master process that performs serialized MPI calls in two threads.
Regards James
Rajeev Thakur wrote:
Actually serialized and multiple thread levels are supported for sock and nemesis. I have updated the README. For other channels such as ssm, which have no thread support, we currently allow single and funneled. Do you need to use serialized with ssm?
Rajeev
-----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of James S Perrin Sent: Friday, August 17, 2007 9:55 AM To: [email protected] Subject: [MPICH2-dev] MPICH2 1.0.6rc1 thread support
Hi, The README states that and serialized and multiple thread levels are only support by the sock channel, but I can configure the ssm channel to use serialized i.e.:
--with-device=ch3:ssm --enable-threads=serialized
where
--with-device=ch3:ssm --enable-threads=multiple
reports a configure error as expected. Is configure correct or the README?
BTW some sort of feature matrix for the various devices and channels would be very helpful.
Regards James -- -------------------------------------------------------------- ---------- James S. Perrin, | email: [email protected] Research Computing Services, | web: www.mc.manchester.ac.uk Kilburn Building, The University, | tel: +44 161 275 6945 Manchester, England. M13 9PL. | fax: +44 161 275 0637 -------------------------------------------------------------- ---------- "The test of intellect is the refusal to belabour the obvious" - Alfred Bester -------------------------------------------------------------- ----------
-- -------------------------------------------------------------- ---------- James S. Perrin, | email: [email protected] Research Computing Services, | web: www.mc.manchester.ac.uk Kilburn Building, The University, | tel: +44 161 275 6945 Manchester, England. M13 9PL. | fax: +44 161 275 0637 -------------------------------------------------------------- ---------- "The test of intellect is the refusal to belabour the obvious" - Alfred Bester -------------------------------------------------------------- ----------
Dear Rajeev, That's great! Thanks very much for your help. Regards James Rajeev Thakur wrote:
James, OK, we will allow thread serialized for ssm and shm channels as well (in 1.0.6). We will also provide a feature matrix. See the example in http://www-unix.mcs.anl.gov/~balaji/support.html .
Regards, Rajeev
-----Original Message----- From: James S Perrin [mailto:[email protected]] Sent: Friday, August 17, 2007 12:03 PM To: Rajeev Thakur Cc: [email protected] Subject: Re: [MPICH2-dev] MPICH2 1.0.6rc1 thread support
Hi,
We want to use ssm as we sometimes have two specific processes per machine that working as a team.
Actaully I think we have been using assuming serialized support at compile time (we're not doing runtime checking of threads_provided, long story don't ask) and it seems to have been working, we hadn't explicitly enabled threading in mpich2 so I gues we've been "lucky" :) Our Master/Slaves app has a threaded Master process that performs serialized MPI calls in two threads.
Regards James
Rajeev Thakur wrote:
Actually serialized and multiple thread levels are supported for sock and nemesis. I have updated the README. For other channels such as ssm, which have no thread support, we currently allow single and funneled. Do you need to use serialized with ssm?
Rajeev
-----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of James S Perrin Sent: Friday, August 17, 2007 9:55 AM To: [email protected] Subject: [MPICH2-dev] MPICH2 1.0.6rc1 thread support
Hi, The README states that and serialized and multiple thread levels are only support by the sock channel, but I can configure the ssm channel to use serialized i.e.:
--with-device=ch3:ssm --enable-threads=serialized
where
--with-device=ch3:ssm --enable-threads=multiple
reports a configure error as expected. Is configure correct or the README?
BTW some sort of feature matrix for the various devices and channels would be very helpful.
Regards James -- -------------------------------------------------------------- ---------- James S. Perrin, | email: [email protected] Research Computing Services, | web: www.mc.manchester.ac.uk Kilburn Building, The University, | tel: +44 161 275 6945 Manchester, England. M13 9PL. | fax: +44 161 275 0637 -------------------------------------------------------------- ---------- "The test of intellect is the refusal to belabour the obvious" - Alfred Bester -------------------------------------------------------------- ----------
-- -------------------------------------------------------------- ---------- James S. Perrin, | email: [email protected] Research Computing Services, | web: www.mc.manchester.ac.uk Kilburn Building, The University, | tel: +44 161 275 6945 Manchester, England. M13 9PL. | fax: +44 161 275 0637 -------------------------------------------------------------- ---------- "The test of intellect is the refusal to belabour the obvious" - Alfred Bester -------------------------------------------------------------- ----------
-- ------------------------------------------------------------------------ James S. Perrin, | email: [email protected] Research Computing Services, | web: www.mc.manchester.ac.uk Kilburn Building, The University, | tel: +44 161 275 6945 Manchester, England. M13 9PL. | fax: +44 161 275 0637 ------------------------------------------------------------------------ "The test of intellect is the refusal to belabour the obvious" - Alfred Bester ------------------------------------------------------------------------
participants (2)
-
James S Perrin -
Rajeev Thakur