[ofa-general] RDMA write completion question
Michael Krause
krause at cup.hp.com
Thu May 24 09:19:19 PDT 2007
At 09:09 AM 5/24/2007, Gleb Natapov wrote:
>On Thu, May 24, 2007 at 08:38:00PM +0530, Devesh Sharma wrote:
> > On 5/24/07, Gleb Natapov <glebn at voltaire.com> wrote:
> > >Hi,
> > >
> > > Does local RDMA write completion guaranties that a data that was
> RDMAed is
> > >already accessible in a destination's host _memory_?
> > Local RDMA write completion guarantees that the data you have RDMAed
> > has been copied into the remote buffer, without any data corruption.
>Is this required by IB spec. How HCA can guaranty that the data is actually
>committed into the memory and not travels through a twisty maze of PCI
>buffers all alike?
There are no guarantees. In fact, data corruption can occur within the CA
as well as via the PCI fabric, etc. There are simply no guarantees. IHV
and chipsets do take steps to prevent corruption at a minimum to at least
detect when it occurs. However, detection and prevention are not always
possible in every design and there is a cost to be paid for either or
both. Technology such as IB does a reasonable job at the fabric level but
has no impact on anything else in the end-to-end data integrity requirements.
Mike
>--
> Gleb.
>_______________________________________________
>general mailing list
>general at lists.openfabrics.org
>http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
>
>To unsubscribe, please visit
>http://openib.org/mailman/listinfo/openib-general
More information about the general
mailing list