[libfabric-users] Transport retry counter exceeded

[Supun Kamburugamuve]

Thanks Sean, I'm using credit messages for explicit flow control at the
application layer. So I think, I'm posting the buffers before the sender
sends the data (I'll double check on this). Is there any other reason that
can cause this?

I've read in RDMAMojo site that this can happen due to some attribute
mismatches. But I'm not sure what they are saying. Also the client and
server are on the same machine.

Supun..

On Tue, Sep 27, 2016 at 1:29 PM, Hefty, Sean <sean.hefty at intel.com> wrote:

> > I've written an server/client using verbs provider. Randomly I get the
> > error.
> >
> > transport retry counter exceeded
> >
> >
> > Sometimes the program works perfectly and sometimes it slows down and
> > gives out the above error. Any particular reason this happens?
>
> I would look at how quickly receive buffers are being re-posted.  If the
> receive queue gets overrun and receives are not re-posted, this could
> occur.  I'm assuming that there's not some hardware or link issue.
>
> - Sean
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/libfabric-users/attachments/20160927/5e0223cb/attachment.html>