On Thu, Aug 18, 2005 at 01:05:49PM -0600, Galen Shipman wrote: > Hello, > > I am using RDMA READ for an internal application and I am seeing > problems with performance on OpenIB. I am not seeing the same > performance problems on like hardware using Mellanox VAPI. Perhaps I am > doing something silly. Attached is a patch to rdma_bw.c where I am > seeing only ~192MB/sec which should be closer to 950MB/sec.
Galen, Normally patches should be attached as "plain text" and not as: [-- Attachment #2: rdma_bw_diff --] [-- Type: application/octet-stream, Encoding: 7bit, Size: 2.4K --] [-- application/octet-stream is unsupported (use 'v' to view this part) --] The patch also has "white space damage" - ie white space changes that don't belong in the patch. Just a distraction. > Any help or ideas are appreciated. I can try it out on my boxes. ISTR I was getting ~750 MB/s on the regular rdma_bw.c code. I agree 192 MB/s seems really low for any PCI-X box. I have to wonder if your version (using reads) is not getting the parallelism that writes could acheive. 192MB/s sounds more like ping-pong with large pages. Could you run the ibv_pingpong test and report what that gets? (It's under .../src/userspace/libibverbs/examples/) grant _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
