I apologize for my dyslectic subject. This should be better. We would
very much like to know if anybody else can reproduce the results? If
you need more info please contact us.
Regards,
Asmund (dyslectic programmer)
Asmund Ostvold wrote:
We discovered a strange problem running OFED; We're not sure if it is a
OFED problem but we post it here anyway.
Short description:
We have a program that allocates a set of buffers with valloc, sends
them with ibv_post_send and free them.
This is run in loop;
We have a "caching"-algorithm so that we register memory only the first
time we come across a buffer address.
We starts getting wrong data for parts of sends after a couple of
iterations
There are a few things worth mentioning:
- We must use valloc; the test works with malloc
- We must have a malloc allocating a too large chunk before starting the
loop (the malloc fails)
We have modified the "rdma_lat.c" program to show the error (attached)
Regards
Asmund
------------------------------------------------------------------------
_______________________________________________
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
_______________________________________________
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
