Okay, I tried Open MPI 1.3.3 and nothing has changed -- OpenIB RDMA isn't used if mpi_leave_pinned is 0.
2009/10/3 Roman Cheplyaka <roman.cheply...@gmail.com>: > Jeff, > yes, I'll try openmpi-1.3.3 on Monday. However, I had look at the svn > code before, and haven't noticed any substantial changes in the code > under the question. > The reason I'm writing to devel@ (and not just reporting a bug) is > that I'm interested in how it is _supposed_ to work. Are there people > who wrote the code (or understand it) on the list? > > On 03/10/2009, Jeff Squyres <jsquy...@cisco.com> wrote: >> On Sep 29, 2009, at 1:36 PM, Roman Cheplyaka wrote: >> >>> Thanks for your explanation, George. However: >>> suppose we have leave_pinned = FALSE. Then we go to >>> mca_mpool_rdma_find. There we try to find suitable registration in the >>> cache. Suppose we cannot (cache's empty). Then NULL registration is >>> returned, BTL is skipped and 0 is returned from mca_pml_ob1_rdma_btls. >>> This way RDMA never happends. >>> >>> (To make it clear: this is not some theoretical situation, this is >>> what I'm actually observing. OpenMPI 1.2.9) Can somebody explain >>> please? >>> >> >> Sorry for the delay in replying. >> >> We actually haven't looked at the 1.2.x code in quite a while -- is it >> possible for you to upgrade to the 1.3 series? There have been *many* >> improvements in the overall code base since the 1.2 series. -- Roman I. Cheplyaka
