[1634] in Release_7.7_team
Re: 8.2.17/8.2.18 patch release
daemon@ATHENA.MIT.EDU (Mike Barker)
Wed Feb 3 17:14:32 1999
Date: Wed, 03 Feb 1999 16:47:58 -0500
To: Greg Hudson <ghudson@MIT.EDU>
From: Mike Barker <mbarker@MIT.EDU>
Cc: f_l@MIT.EDU, network@MIT.EDU, release-team@MIT.EDU
In-Reply-To: <199902022358.SAA15033@small-gods.mit.edu>
Response Summary:
1. We are NOT going to put your request in 8.2.17/8.2.18.
2. We are working on an alternative to help with your class lockers. There
are two questions: 1) is it sufficient to always use the system defaults for
rsize and wsize or do you need to be able to adjust these? And 2) is there
a location in your class locker(s) for a special program? Where?
Background:
1. I've talked with Bill Cattey, Greg Hudson, Tom Coppeto, and Dan Winship
about possible approaches.
2. The current rsize and wsize entries in attach.conf are thought to be an
earlier response to a complaint about using the defaults in this same
situation. It is believed that using smaller packet sizes provides improved
behavior depending on network traffic conditions and topology, so we would
prefer NOT to change this without doing significant testing across the
various NFS sites and their usage.
3. It is too late to make a modification with this kind of potential for
widespread impacts in 8.2.17.
4. As an alternative, Dan Winship has prepared a "special" version of
attach which does nothing with rsize and wsize entries in attach.conf or
anywhere else. I.e., it will always use the system defaults for these two
options. He says he can also provide a version that allows the user to
enter those two options on the command line.
I thought we would provide this special version either in a small locker or
in the course locker. The students could then be instructed to run this
special version instead of the normal attach.
5. Just to document, attach does accept command-line mount options.
However, if there are entries in attach.conf, they override these options.
This is intentional to avoid cases such as the system admin limiting suid in
attach.conf, and the user simply overriding that via the command line.
mike
At 06:58 PM 2/2/99 -0500, Greg Hudson wrote:
:)Last time this came up someone grumbled about having a debate in view
:)of the people making the request. So I'm quoting this request for
:)release-team (which didn't get it) and including the IS mailing lists
:)which did get it. My take on the conversation is not the same as
:)Tom's; I thought all of the responses were not in favor of making the
:)change he requested (excepting my own, which was noncommittal).
:)
:)I certainly don't want to put this into 8.2.17, which is already too
:)complicated. It's not clear to me that we want to make his change at
:)all, or what forum it is best to make that decision in. (Certainly I
:)can't make it by myself.) Given Grayson's position, I'd prefer if
:)someone at Mike or Bill's level sent a response.
:)
:)> Message-Id: <199902022247.RAA25965@m2-032-1.mit.edu>
:)> To: Greg Hudson <ghudson@MIT.EDU>
:)> Cc: f_l@MIT.EDU, network@MIT.EDU, 11.520staff@MIT.EDU, phils@MIT.EDU,
:)> karen@MIT.EDU
:)> Subject: Re: 8.2.17/8.2.18 patch release
:)> In-Reply-To: Your message of "Tue, 02 Feb 1999 14:07:52 EST."
:)> <199902021907.OAA06613@small-gods.mit.edu>
:)> Date: Tue, 02 Feb 1999 17:47:03 EST
:)> From: "Thomas H. Grayson" <thg@MIT.EDU>
:)>
:)> Greg,
:)>
:)> I have an important request for the new Athena release, if it is
:)> not already in there (I didn't see it in your list). Please
:)> change the standard Athena file /etc/athena/attach.conf to remove
:)> the small nfs rsize and wsize parameters (currently 1024).
:)> Allowing 'attach' to use the NFS mount defaults greatly improves
:)> NFS performance. Our class 11.520/11.188/11.523/11.524 this term
:)> will be using ArcView and large GIS datasets which reside in
:)> NFS-based lockers. We lose a lot of time at the beginning of
:)> weekly labs when students wait for ArcView, a 10.7 MB executable,
:)> to launch over the network. Unfortunately, hacking the
:)> attach.conf file on a public workstation appears to get undone
:)> overnight. Hence, I'd really like to see this in the release.
:)> There was some debate a couple of months ago on this subject, and
:)> there appeared to be agreement that this was a good idea.
:)>
:)> Here's a sample of a corrected attach.conf file:
:)>
:)> #
:)> # Attach configuration file
:)> #
:)> # $Id: attach.conf,v 1.4 1997/05/08 07:30:30 ghudson Exp $
:)> #
:)> aklog /bin/athena/aklog
:)> # The order of the nomountpoint and mountpoint lines below is important.
:)> nomountpoint ^/afs ^/bin ^/core ^/dev ^/devices ^/etc ^/export
:)> nomountpoint ^/home ^/hsfsboot ^/kadb ^/kernel ^/lib
^/lost+found
:)> nomountpoint ^/opt ^/os ^/platform ^/proc ^/sbin ^/srvd ^/tmp
:)> nomountpoint ^/ufsboot ^/usr ^/var ^/vol ^/mit/[^/]*/. ^/\.
:)> mountpoint /mit/[^/]*
:)> nomountpoint /mit
:)> nosetuid .
:)> options {nfs}:.* soft
:)>
:)>
:)> Tom Grayson
:)> GIS/Database Specialist
:)> Dept. of Urban Studies and Planning