netdev-ow...@vger.kernel.org wrote on 09/27/2010 10:54:44 AM:
> From: David Miller <da...@davemloft.net>
> To: David Stevens/Beaverton/i...@ibmus
> Cc: c...@linux.com, linux-rdma@vger.kernel.org,
> net...@vger.kernel.org, netdev-ow...@vger.kernel.org, r...@rincon.com
> Date: 09/27/2010 10:54 AM
> Subject: Re: igmp: Allow mininum interval specification for igmp timers.
> Sent by: netdev-ow...@vger.kernel.org
>
> From: David Stevens <dlstev...@us.ibm.com>
> Date: Mon, 27 Sep 2010 10:41:20 -0700
>
> > I don't know if you saw the more extended discussion we
> > had on this or not, but I think while this would help for IB,
> > it's not appropriate in general. These can in fact be "0" per
> > RFC which is worst case for IB if there is a delay for being
> > able to use the group, and the newer IGMPv3 standard has shortened
> > the max interval from 10sec in v2 to 1 sec.
>
> I did see the extended discussion, and it was interesting :-)
>
> But that mainly focused on the second patch, which I appropriately
> marked as needing changes in patchwork.
OK, I'm not sure I've seen them all; haven't caught up on
e-mail yet.
> This patch on the other hand is attacking a different problem,
> namely avoiding the worst cases caused by the randomization we
> do for the timer.
I think the multiples are to allow for drops and the
randomization is to prevent storms. As far as IGMP is concerned,
it's perfectly fine to send them back-to-back, since drops are
not necessarily time periods of network outage (as with IB) but
rather transient queue overflows where even the short delay of
a "0" timer but still having protocol and packet transmit delay
would be fine.
> With bad luck this thing times out way too fast because the total of
> all of the randomized intervals can end up being very small, and I
> think we should fix that independently of the other issues hit by the
> IB folks.
>
> Don't you agree?
If you mean enforcing a minimum spacing higher than a "0" timer,
I don't know that it's an issue for other network types. According to
IGMPv3, all of them (3 total) on average would be sent in 1 sec, but
it also isn't fatal to drop all of them. To the extent that 1 sec is
"small," it is intentional.
I'll try digging out the particular patch and comment. I'm not
sure many of these tweaks would necessarily hurt other network types
but I think the current code also isn't a problem for anything but IB,
and that issue can be fixed within IB.
+-DLS
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
