[1324] in SIPB-AFS-requests

home help back first fref pref prev next nref lref last post

Re: server migration

daemon@ATHENA.MIT.EDU (mhpower@MIT.EDU)
Sun Apr 17 00:59:20 1994

From: mhpower@MIT.EDU
To: ghudson@MIT.EDU
Cc: sorokin@MIT.EDU, sipb-afsreq@MIT.EDU, charon-maintainers@MIT.EDU,
        rtfm-maintainers@MIT.EDU, webmaster@MIT.EDU, usenet@MIT.EDU
In-Reply-To: "[1320] in SIPB-AFS-requests", "[0750] in Charon_Maintainers_Archive", "[1359] in RTFM_Maintainers_Archive", "[1218] in Webmaster", "[28321] in Usenet_Meeting"
Date: Sun, 17 Apr 94 00:58:17 EDT

>                                                   ... These will
>break down as follows:
>
>	AFS:		Two maxines

I'm not sure this is necessary. We originally added a second server to
the sipb cell because we got rosebud, which happened to have lots of
free space on its internal disks, and we wanted that disk space to be
useful. Presumably the new arrangement would involve only external
disks (sure, we could carve out a small /vicepX partition from an
rz25, but that's mostly irrelevant). Do we really have a justification
for having two afs servers any more? I think it might work to install
a 3 Gb disk as /dev/rz4c on ronald-ann, vos move all the rosebud
volumes onto it, and then go drop rosebud off the roof. I think this
would be much simpler in terms of both the changeover (i.e., there
would be no afs outage, at all) and long-term sipb afs maintenance
(e.g., no possible inconsistency in databases, binaries, users, etc.)

>Currently, I favor using the SS20 for beacon.

Another reason is that it seems entirely possible that a lot of the
problems with stuck nntp connections are due to Ultrix kernel bugs.

>                                           ... The Web server should
>do significantly better on a Maxine with 40MB of memory than it does
>now on charon, ...

Right, but the growth of www is much faster than any other sipb
service. I don't think putting one Maxine there works either. One
possibility is to run the www server on the same machine as the rtfm
services. This assumes that an ss20 is more than "twice as good" as a
Maxine (in terms of whatever resources the services need). I'm not
sure whether that's really true. I think a better solution that's
still feasible is to assign two Maxines to the www server, and letting
the webmasters decide whether they want to have some files/URLs on one
machine and some on the other, or whether they want duplication for
load balancing (possibly something like is done on www.ncsa.uiuc.edu).
This could involve either the Maxine tagged for afs, or else yaz.

>Several other services such as bitnet and gopher will probably run on
>the same machine as discuss or WWW.

I think it doesn't particularly matter which machines these run on, so
we might as well put them wherever it'll be simplest in terms of
software changes and cname assignments (i.e., it will save a lot of
work/trouble if discuss is on a machine with the names charon and
bloom-picayune, bitnet is also there, the names sipb and rtfm are both
on the same other machine, etc.) Also, we need some place to put irc.

Matt

home help back first fref pref prev next nref lref last post