I was thinking of trying to run Cactus on supermic and attempting to use the mics. I'm not expecting to receive any speedup from doing so, but I'm going to try.
One thing I'm wondering is whether I would receive a speedup if the mic nodes had a different number of gridpoints than the CPUs (again, not overly optimistic). Is there a way to tell a load balancer to give half the nodes a larger share of gridpoints? If there isn't, how hard would it be to create such a load balancer?
--Steve
Steve
No, there is unfortunately no such mechanism yet. We have two "new" (not yet working) load balancers in Carpet that would have this functionality, though. Debugging these might be the easiest way to go.
-erik
On Wed, Mar 15, 2017 at 4:57 PM, Steven R. Brandt sbrandt@cct.lsu.edu wrote:
I was thinking of trying to run Cactus on supermic and attempting to use the mics. I'm not expecting to receive any speedup from doing so, but I'm going to try.
One thing I'm wondering is whether I would receive a speedup if the mic nodes had a different number of gridpoints than the CPUs (again, not overly optimistic). Is there a way to tell a load balancer to give half the nodes a larger share of gridpoints? If there isn't, how hard would it be to create such a load balancer?
--Steve
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
Are these not-yet-working balancers almost-working? Where are they?
--Steve
On 03/15/2017 05:36 PM, Erik Schnetter wrote:
Steve
No, there is unfortunately no such mechanism yet. We have two "new" (not yet working) load balancers in Carpet that would have this functionality, though. Debugging these might be the easiest way to go.
-erik
On Wed, Mar 15, 2017 at 4:57 PM, Steven R. Brandt <sbrandt@cct.lsu.edu mailto:sbrandt@cct.lsu.edu> wrote:
I was thinking of trying to run Cactus on supermic and attempting to use the mics. I'm not expecting to receive any speedup from doing so, but I'm going to try. One thing I'm wondering is whether I would receive a speedup if the mic nodes had a different number of gridpoints than the CPUs (again, not overly optimistic). Is there a way to tell a load balancer to give half the nodes a larger share of gridpoints? If there isn't, how hard would it be to create such a load balancer? --Steve _______________________________________________ Users mailing list Users@einsteintoolkit.org <mailto:Users@einsteintoolkit.org> http://lists.einsteintoolkit.org/mailman/listinfo/users <http://lists.einsteintoolkit.org/mailman/listinfo/users>-- Erik Schnetter <schnetter@cct.lsu.edu mailto:schnetter@cct.lsu.edu> http://www.perimeterinstitute.ca/personal/eschnetter/
Steve
Yes, they are almost working. Each a blocked by a single bug that we haven't found yet. (Okay, I'm not making a promise about the "single" part.) They are part of Carpet, and can be enabled by setting the parameter Carpet::processor_topology. The new ones are "recursive" and "balanced", written by Peter Diener and myself, respectively.
-erik
On Thu, Mar 16, 2017 at 9:58 AM, Steven R. Brandt sbrandt@cct.lsu.edu wrote:
Are these not-yet-working balancers almost-working? Where are they?
--Steve
On 03/15/2017 05:36 PM, Erik Schnetter wrote:
Steve
No, there is unfortunately no such mechanism yet. We have two "new" (not yet working) load balancers in Carpet that would have this functionality, though. Debugging these might be the easiest way to go.
-erik
On Wed, Mar 15, 2017 at 4:57 PM, Steven R. Brandt sbrandt@cct.lsu.edu wrote:
I was thinking of trying to run Cactus on supermic and attempting to use the mics. I'm not expecting to receive any speedup from doing so, but I'm going to try.
One thing I'm wondering is whether I would receive a speedup if the mic nodes had a different number of gridpoints than the CPUs (again, not overly optimistic). Is there a way to tell a load balancer to give half the nodes a larger share of gridpoints? If there isn't, how hard would it be to create such a load balancer?
--Steve
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
-- Erik Schnetter schnetter@cct.lsu.edu http://www.perimeterinstitute.ca/personal/eschnetter/
users@lists.einsteintoolkit.org
-
Erik Schnetter -
Steven R. Brandt