[47488] in SIPB-AFS-requests
Re: sipb cell AFS server tuning
daemon@ATHENA.MIT.EDU (Mitchell E Berger)
Tue Jun 18 10:28:39 2013
From mitchb@MIT.EDU Tue Jun 18 14:28:39 2013
Return-Path: <mitchb@MIT.EDU>
Delivered-To: sipb-afsreq-mtg@CHARON.mit.edu
Received: (qmail 7256 invoked from network); 18 Jun 2013 14:28:39 -0000
Received: from mailhub-auth-1.mit.edu (18.9.21.35)
by charon.mit.edu with SMTP; 18 Jun 2013 14:28:39 -0000
Received: from outgoing.mit.edu (outgoing-auth-1.mit.edu [18.9.28.11])
by mailhub-auth-1.mit.edu (8.13.8/8.9.2) with ESMTP id r5IESWtq018162;
Tue, 18 Jun 2013 10:28:32 -0400
Received: from byte-me.mit.edu (byte-me.mit.edu [18.187.2.49])
(authenticated bits=56)
(User authenticated as mitchb@ATHENA.MIT.EDU)
by outgoing.mit.edu (8.13.8/8.12.4) with ESMTP id r5IESUVY003850
(version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NOT);
Tue, 18 Jun 2013 10:28:31 -0400
Received: (from mitchb@localhost) by byte-me.mit.edu (8.12.9.20060308)
id r5IESTKC028318; Tue, 18 Jun 2013 10:28:29 -0400 (EDT)
Date: Tue, 18 Jun 2013 10:28:29 -0400 (EDT)
From: Mitchell E Berger <mitchb@MIT.EDU>
To: Jonathon Weiss <jweiss@MIT.EDU>
cc: sipb-afsreq@MIT.EDU
Subject: Re: sipb cell AFS server tuning
In-Reply-To: <201306181328.r5IDSjpd012090@outgoing.mit.edu>
Message-ID: <Pine.GSO.4.64L.1306181023060.8087@byte-me.mit.edu>
References: <201306181328.r5IDSjpd012090@outgoing.mit.edu>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII; format=flowed
I think the reason we didn't disable weekly restarts long ago was
basically from an understanding that it used to be needed to cope
with memory leaks and such bugs in older versions of the software,
which had become less prevalent as development went on, and because
reynelda was running a fairly old version of everything, we might
be setting ourselves up for a crash if we stopped doing that.
Given that reynelda's coming up to the chopping block soon, I think
we might as well leave the restarts going while old-rc is still in
service and discontinue them as soon as the rc upgrade occurs. (It
seldom inconveniences anyone these days anyhow, as from the time all
the volumes were on speedy Linux servers, the sipb cell restart has
taken under a minute if I recall correctly.)
I don't have an intuitive feel for the settings of the other parameters,
but the proposed changes sound reasonable to me.
Mitch
On Tue, 18 Jun 2013, Jonathon Weiss wrote:
>
>
> I just noticed a few things about the sipb AFS cell.
>
> * it looks like the fileserver is using the arguments that ops
> selected for the athena cell years ago.
>
> * ops recently changed those to increas the number of threads
> ("-p 36" --> "-p 64") and the number of callback slots ("" -->
> "-cb 256000"), and there are indications that the sipb cell
> would benefit from at least the second of these (and possibly
> the first, but that's less clear).
>
> * the sipb cell still restarts weekly. The athena cell hasn't
> done this in a long time.
>
> I'd like to see us update both fileserver arguments, and disable the
> weekly restart. That said, I don't want to interfere with the server
> upgrades Mitch and Ben are working on. Do folks agree in concept with
> these changes, and if so, have any thoughts on coordinateion? (I could
> certainly stageteh arg changes in BosConfig let one last weekly restart
> happen, and then disable that.)
>
> Jonathon
>
