| Message ID | 20110706151116.79c187ea.weiny2@llnl.gov (mailing list archive) |
|---|---|
| State | Accepted |
| Delegated to: | Alex Netes |
| Headers | show |
Ira,
On 7/8/2011 12:06 PM, Ira Weiny wrote:
> However, when we first put the PerfMgr in OpenSM there was some concern for developers who may be using OpenSM in an embedded environment.
Do you recall the specific concerns ? I forget and would rather not dig
them out unless really necessary...
-- Hal
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Ira, On 10:13 Fri 08 Jul , Ira Weiny wrote: > On Fri, 8 Jul 2011 09:23:50 -0700 > Hal Rosenstock <hal@dev.mellanox.co.il> wrote: > > > Ira, > > > > On 7/8/2011 12:06 PM, Ira Weiny wrote: > > > However, when we first put the PerfMgr in OpenSM there was some concern for developers who may be using OpenSM in an embedded environment. > > > > Do you recall the specific concerns ? I forget and would rather not dig > > them out unless really necessary... > > If I recall there was no "specific" concern just that someone _may_ do this. > > Also I think there was some concern about the PerfMgr taking too much processing time with it's threads. But I think that is configurable by the number of SMP's on the wire and turning it off if one prefers a different model. > PerfMgr adds ~200K to opensm binary (34K when stripped). During runtime PerfMgr adds ~270K to opensm memory allocation, when running with PerfMgr disabled. During the opensm initialization, osm_perfmgr_init() and osm_perfmgr_bind() are called even if PerfMgr is disabled. I guess it's needed, if we want to enable PerfMgr later on the fly. I'm not sure though, how calling to osm_perfmgr_bind() affects performance. One more thing, perfmgr_sweep() is called each (pm->sweep_time_s * 1000) no matter if PerfMgr is enabled or disabled. Though it does nothing, I think it's better to execute perfmgr_sweep() only when PerfMgr is enabled. -- Alex -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, 10 Jul 2011 01:54:18 -0700 Alex Netes <alexne@mellanox.com> wrote: > Hi Ira, > > On 10:13 Fri 08 Jul , Ira Weiny wrote: > > On Fri, 8 Jul 2011 09:23:50 -0700 > > Hal Rosenstock <hal@dev.mellanox.co.il> wrote: > > > > > Ira, > > > > > > On 7/8/2011 12:06 PM, Ira Weiny wrote: > > > > However, when we first put the PerfMgr in OpenSM there was some concern for developers who may be using OpenSM in an embedded environment. > > > > > > Do you recall the specific concerns ? I forget and would rather not dig > > > them out unless really necessary... > > > > If I recall there was no "specific" concern just that someone _may_ do this. > > > > Also I think there was some concern about the PerfMgr taking too much processing time with it's threads. But I think that is configurable by the number of SMP's on the wire and turning it off if one prefers a different model. > > > > PerfMgr adds ~200K to opensm binary (34K when stripped). > During runtime PerfMgr adds ~270K to opensm memory allocation, when running > with PerfMgr disabled. > > During the opensm initialization, osm_perfmgr_init() and osm_perfmgr_bind() > are called even if PerfMgr is disabled. I guess it's needed, if we want to > enable PerfMgr later on the fly. Correct. > I'm not sure though, how calling to osm_perfmgr_bind() affects performance. I don't think it should affects performance. > > One more thing, perfmgr_sweep() is called each (pm->sweep_time_s * 1000) no > matter if PerfMgr is enabled or disabled. Though it does nothing, I think it's > better to execute perfmgr_sweep() only when PerfMgr is enabled. Subsequent patch on it's way. Ira > > > -- Alex
Hi Ira, On 15:11 Wed 06 Jul , Ira Weiny wrote: > This should at least be compiled in by default. > > > Signed-off-by: Ira Weiny <weiny2@llnl.gov> > --- Applied, thanks. -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/config/osmvsel.m4 b/config/osmvsel.m4 index c24930b..2c91f63 100644 --- a/config/osmvsel.m4 +++ b/config/osmvsel.m4 @@ -206,12 +206,12 @@ AC_DEFUN([OPENIB_OSM_PERF_MGR_SEL], [ dnl enable the perf-mgr AC_ARG_ENABLE(perf-mgr, -[ --enable-perf-mgr Enable the performance manager (default no)], +[ --enable-perf-mgr Enable the performance manager (default yes)], [case $enableval in yes) perf_mgr=yes ;; no) perf_mgr=no ;; esac], - perf_mgr=no) + perf_mgr=yes) AC_ARG_ENABLE(perf-mgr-profile, [ --enable-perf-mgr-profile Enable the performance manager profiling (default no)], [case $enableval in
This should at least be compiled in by default. Signed-off-by: Ira Weiny <weiny2@llnl.gov> --- config/osmvsel.m4 | 4 ++-- 1 files changed, 2 insertions(+), 2 deletions(-)