[mvapich-discuss] RE: data placement in-order guarantee ?

[Viral Mehta]

Hi,
You can either do a "simple send" after RDMA write or you can also poll on last byte... While I was working on it, I had done and I never ran into problem with either of the method.

Someone also quoted texts from Infiniband Spec, but I would also suggest to read iWARP spec if you are going to support both of them.... For example, in case you decide to implement rdma_write_atomic then,
1. infiniband supports rdma_write_atomic and generates completion on both the sides..
2. iwarp does not support rdma_write_atomic and thus middle level libraries (probably like udapl) will actually do rdma_write operation and will generate completion on one side only...

Hope it may help :)

Thanks,
Viral

-----Original Message-----
From: linux-rdma-owner at vger.kernel.org [mailto:linux-rdma-owner at vger.kernel.org] On Behalf Of yogeshwar sonawane
Sent: Tuesday, February 02, 2010 8:44 PM
To: linux-rdma at vger.kernel.org; mvapich discuss
Cc: yogyas at gmail.com
Subject: data placement in-order guarantee ?

Hi all,

Following question can be of naïve nature. Kindly, help me to clarify
the points.
Thanks in Advance.

Consider an application on top of uverbs / udapl library and wants to
use only rdma write operations.
While using RDMA write operations, one has to indicate the responder
side(receiving side) somehow that the data transfer is complete. This
can be achieved simply by doing an extra rdma write (sort of control
write) conveying successful completion of DTO. Instead of this, can
programmer poll/check on the last data (int or byte) assuming in-order
placement of data in the receive buffer.
Can such behavior of last byte written last, assumed ? If application
programmer can rely on such behavior, then this asks for all networks
to guarantee such behavior.

It will be helpful if some information is shared as how mvapich2
handles the situation regarding above point (while using rdma
operations) ?

I thought that i should mail to both the mailing lists as topic covers
both domains.

Thanks and Regards,
Yogeshwar
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

______________________________________________________________________

This Email may contain confidential or privileged information for the intended recipient (s) If you are not the intended recipient, please do not use or disseminate the information, notify the sender and delete it from your system.

______________________________________________________________________