MIME-Version: 1.0
In-Reply-To: <rt-3.8.6-2416-1409709387-330.37059-5-0@isc.org>
Content-Disposition: inline
References: <RT-Ticket-37059@isc.org> <20140903000601.GA9456@uow.edu.au> <rt-3.8.6-2416-1409702768-487.37059-3-0@isc.org> <20140903015618.CA5801E3F8E0@rock.dv.isc.org> <rt-3.8.6-2416-1409709387-330.37059-5-0@isc.org>
Message-ID: <20140903024606.GA1885@uow.edu.au>
Content-Type: text/plain; charset="utf-8"
Subject: Re: [ISC-Bugs #37059] runtime error
Date: Wed, 3 Sep 2014 12:46:06 +1000
To: Mark Andrews via RT <bind9-bugs@isc.org>
From: "Peter D. Gray" <pdg@uow.edu.au>
RT-Message-ID: <rt-3.8.6-2416-1409712371-1098.37059-0-0@isc.org>
Content-Length: 1041

Hi Mark, thank you for the quick reply.

I just wanted you to know about the problem.

On Wed, Sep 03, 2014 at 01:56:28AM +0000, Mark Andrews via RT wrote:
> 
> Well 16M is less than a second's worth of traffic these days.
>

Well yes, but in what situation would not a single thread call recv
for a full second. If thats happening, you need more threads.

> You may want to tune -U #listeners down from the default and/or the
> number if cpus (-n) which reduces -U.
> 
> That said the ability to set this size at runtime would be useful.

Agreed, but it sounds like creeping featurism.

You could loop, trying smaller values until you found one that worked,
but that sounds like overkill to me.

The maximum allowable value for rcvbufsiz on solaris by default is
2M for UDP and 1M for TCP. I have no idea how to find that out at compile time
or at run time for that matter.

The maximum value for the above can be set as a kernel tunable.

I have set the smaller of the 2 recbufsiz sizes to 256k for my build
and used that.

Regards,
pdg