[libfabric-users] connection-less send/recv with verbs

[Ilango, Arun]

>As far as I understand, fabtests use a spinning wait on top of asynchronous send/recv communication, that should 
>definitively not be influenced by the relative starting order.

Actually, you're right. The retry-wait mechanism in the provider or hardware should take care of the re-sending the request again in case it doesn't get a response at first. And this works for me in my setup with ofi_rxm, and sockets providers but not with verbs. There may be some issue in the verbs/RDM implementation.

> Indeed, with this command line the “verbs;ofi_rxm” shows up, maybe it is just matter of some limit on the number of 
>results showed by fi_info.

AFAIK there isn't any limit to the number of results that show up in fi_info.

-Arun.

-----Original Message-----
From: Maurizio Drocco [mailto:drocco at di.unito.it] 
Sent: Monday, July 17, 2017 12:38 PM
To: Ilango, Arun <arun.ilango at intel.com>
Cc: libfabric-users at lists.openfabrics.org
Subject: Re: [libfabric-users] connection-less send/recv with verbs

Thank you Arun,

> The server has to be started before the client and it waits for the first message from client.
> 
> The handshake lets the server and the client know each other's fabric addresses which is then used in the data transfer calls.

I understand how the handshake works, but I do not get the requirement of starting the server before the client, in particular in case of connection-less communication.
As far as I understand, fabtests use a spinning wait on top of asynchronous send/recv communication, that should definitively not be influenced by the relative starting order.

Anyway, I think we should consider as resolved the issue I posed.

Conversely, I am collecting a set of behaviors that look to me like not fully matching the documentation.
What is the proper path to report them? Maybe mailing to the ofiwg mailing list?

> FI_LOG_LEVEL=debug FI_LOG_PROV="verbs,ofi_rxm" fi_info -p verbs

Indeed, with this command line the “verbs;ofi_rxm” shows up, maybe it is just matter of some limit on the number of results showed by fi_info.
I will give it a try.

Thank you again for your time.

Maurizio