[erlang-questions] disconnected nodes

[Ahmed Omar]

Hi Ignas,

Yes I meant the "volume" or (number*size of the messages sent).
In our case the messages were plenty of XMPP messages, which could be
large.

In your case, maybe it's different.

P.S: Network utilisation might not be enough to judge, there might be some
kind of microbrusts of traffic as described in this informative blog post
by Scott Lystig Fritchie
http://www.snookles.com/slf-blog/2012/01/05/tcp-incast-what-is-it/

Best Regards,
- Ahmed Omar
http://about.me/spawn.think/


On Tue, Feb 18, 2014 at 3:09 PM, Ignas Vyšniauskas <baliulia@REDACTED>wrote:

> Hi again,
>
> Thanks for the response.
>
> On 02/17/2014 11:23 AM, Ahmed Omar wrote:
> > The reason for the later incidents, was usually high traffic between
> >  the nodes. That traffic could cause distribution buffer to reach
> > limits and suspend processes trying to communicate with other
> > processes on other nodes.
> > (http://erlang.org/doc/man/erlang.html#system_info_dist_buf_busy_limit)
> >
> > One quick work around for that would be to increase the limit using
> > +zdbbl flag on startup. A more permanent solution would be to look
> > into the traffic between the nodes and optimise there.
>
> By high traffic do you mean amount of messages or actual volume?
>
> AFAIK, +zdbbl is more of a possible cure for volume and battling things
> like TCP incast. However, in our case the network utilization does not
> go over ~20%, so I don't think TCP incast is playing role here. The
> number of messages is quite large, though.
>
> I bumped it to 32MB in any case,  can't hurt.
>
> --
> Ignas
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20140218/0af95436/attachment.htm>