linux-rdma.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* rdma_connect() timeout question
@ 2012-08-16 19:01 Venkat Venkatsubra
       [not found] ` <502D43A4.1090708-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org>
  0 siblings, 1 reply; 3+ messages in thread
From: Venkat Venkatsubra @ 2012-08-16 19:01 UTC (permalink / raw)
  To: linux-rdma; +Cc: Michael Nowak

Hello,

Is there a way to tune rdma_connect() timeout to a lower value ?
With CMA_CM_RESPONSE_TIMEOUT as 20 and CMA_MAX_CM_RETRIES 15,
it takes about a minute to timeout. We would like to tune it in our case 
to a much lower value.

The scenario is RDS trying to reconnect when a bonding interface 
notifies us of a link fail over.
In our setup the gratuitous arp doesn't arrive sometimes to update the 
arp entry with the new GID.
The address resolution returns the old GID. Route resolution also 
happens to return the old path in this failure case.
(Other times when route solution fails it is not a problem. The 
reconnect attempts eventually go through in those cases.)
And the rdma_connect() gets stuck. A timeout in 1 minute and a 
subsequent retry of address/route-resolution/rdma_connect then succeeds.
Because by that time the arp entry is updated.

Can cma_response_timeout and CMA_MAX_CM_RETRIES be made tunables ?
Even better would have been if the caller (RDS in this case) could have 
passed them as connection setup parameters. ;-)
That way we don't have to disturb the setting for non-RDS traffic.

Venkat


--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

^ permalink raw reply	[flat|nested] 3+ messages in thread

* RE: rdma_connect() timeout question
       [not found] ` <502D43A4.1090708-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org>
@ 2012-08-16 19:40   ` Hefty, Sean
       [not found]     ` <1828884A29C6694DAF28B7E6B8A8237346A899E6-P5GAC/sN6hmkrb+BlOpmy7fspsVTdybXVpNB7YpNyf8@public.gmane.org>
  0 siblings, 1 reply; 3+ messages in thread
From: Hefty, Sean @ 2012-08-16 19:40 UTC (permalink / raw)
  To: Venkat Venkatsubra, linux-rdma; +Cc: Michael Nowak, Steve Wise, Sharp, Robert O

> Is there a way to tune rdma_connect() timeout to a lower value ?

I don't believe so.

You could abort the connection by destroying the id after a specified amount of time had expired.

> Can cma_response_timeout and CMA_MAX_CM_RETRIES be made tunables ?
> Even better would have been if the caller (RDS in this case) could have
> passed them as connection setup parameters. ;-)
> That way we don't have to disturb the setting for non-RDS traffic.

A per rdma_cm_id option for this makes sense to me. 

iWarp developers, would timeout/retry values be usable for iwarp?
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: rdma_connect() timeout question
       [not found]     ` <1828884A29C6694DAF28B7E6B8A8237346A899E6-P5GAC/sN6hmkrb+BlOpmy7fspsVTdybXVpNB7YpNyf8@public.gmane.org>
@ 2012-08-16 19:57       ` Steve Wise
  0 siblings, 0 replies; 3+ messages in thread
From: Steve Wise @ 2012-08-16 19:57 UTC (permalink / raw)
  To: Hefty, Sean
  Cc: Venkat Venkatsubra, linux-rdma, Michael Nowak, Sharp, Robert O

On 8/16/2012 2:40 PM, Hefty, Sean wrote:
>> Is there a way to tune rdma_connect() timeout to a lower value ?
> I don't believe so.
>
> You could abort the connection by destroying the id after a specified amount of time had expired.
>
>> Can cma_response_timeout and CMA_MAX_CM_RETRIES be made tunables ?
>> Even better would have been if the caller (RDS in this case) could have
>> passed them as connection setup parameters. ;-)
>> That way we don't have to disturb the setting for non-RDS traffic.
> A per rdma_cm_id option for this makes sense to me.
>
> iWarp developers, would timeout/retry values be usable for iwarp?
I think this would be useful for the iwarp side too.  Not sure yet 
exactly how it would apply for cxgb3/4 though...

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2012-08-16 19:57 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2012-08-16 19:01 rdma_connect() timeout question Venkat Venkatsubra
     [not found] ` <502D43A4.1090708-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org>
2012-08-16 19:40   ` Hefty, Sean
     [not found]     ` <1828884A29C6694DAF28B7E6B8A8237346A899E6-P5GAC/sN6hmkrb+BlOpmy7fspsVTdybXVpNB7YpNyf8@public.gmane.org>
2012-08-16 19:57       ` Steve Wise

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).