[1259] in SIPB-AFS-requests
Server Hardware proposal
daemon@ATHENA.MIT.EDU (yandros@MIT.EDU)
Mon Feb 28 23:25:18 1994
From: yandros@MIT.EDU
Date: Tue, 1 Mar 94 04:24:32 GMT
To: jis@MIT.EDU
Cc: yandros@MIT.EDU, sipb-staff@MIT.EDU
Cc: rtfm-maintainers@MIT.EDU, charon-maintainers@MIT.EDU, sipb-ec@MIT.EDU,
usenet@MIT.EDU, sipb-afsreq@MIT.EDU
(~yandros/project/sipb/server-prop.txt)
Before:
~~~~~~~
AFS
===
RS/6000 --- 2.5 Gig
DS5k/25 --- 3 Gig
News
====
DS5k/200 news 2.6 Gig
DS5k/25 transport 1 Gig
DS5k/25 slave/test 0 Gig 40 M
Meta
====
DS5k/25 rtfm, discuss 1 Gig
DS3100 WWW, discuss .3 Gig
After:
~~~~~~
AFS
===
DS5k/25 --- 3 Gig
DS5k/25 --- 3 Gig
News
====
SS10(2) news 3 Gig 80 M
SS10(3) transport 1 Gig 40 M
DS5k/25 slave/test 1 Gig
Meta
====
DS5k/25 discuss 1 Gig
SS10(1) rtfm 1 Gig 64 M
DS5k/25 WWW & NFS 1 Gig 40 M
Delta's:
~~~~~~
o Machines
- 3 machines are returned: an RS/6000, a DS3100, and a DS5k/200
- 4 machines come in: 3 SS10's (ranked in order of necessity, in
case it's necessary to make one of them a DS5k/25 instead of an
SS10) and one DS5k/25.
o Memory
- one of the DS5k/25's currently has 40M of memory. This memory
would be in the machine used for WWW and NFS service (the WWW
server is somewhat memory-intensive)
- we estimate that the news reader server would need 80M total
- we estimate that the news transport server would need 40M total
- we estimate that the rtfm server would need 64M total
o Disk
- Added:
2 disks replace the internal disks on the RS6k AFS server.
1 disk is added to the news reader, for normal growth
1 disk is added for NFS service to Resnet
1 disk is added to consolidate centralized Discuss service
(replacing the RZ55 currently on charon)
1 disk will allow us to test news slave service
- Returned:
3 RZ55's will be returned.
2 .7 Gig disks will be returned as part of the RS6k AFS server.
Concerns addressed:
~~~~~~~~~~~~~~~~~~~
o RTFM becomes unhosed
RTFM is currently nigh-unusable. We've had to disable a very
useful, very important service (WAIS access) in order to keep it
afloat now.
o News becomes unhosed and can grow.
This configuration, combined with the code changes we expect to have
by then, should enable news service to continue to be usable for
a long while, while we test and evaluate slave service.
One additional reason for requesting two SS10's for news service is
so that our production servers can both run the same OS. (SunOS is
a major win for our server implementation).
o Discuss becomes faster
This proposal adds a mostly-dedicated machine for discuss service
(of the type IS is currently using), which should help to solve some
of the slowness and reliability concerns that are starting to crop
up. (especially in that we can remove discuss service from the
over-net-loaded rtfm).
o Added NFS server to support ResNet (Linux, NetBSD, DOS/Windows, OS/2, etc.)
We *need* an NFS server to support ResNet adequately; currently we
are `sliming' NFS space from two watchmaker machines and still
letting services fall down due to lack of usable disk space we can
export reliably.
o Disk to test news slave service
The additional disk will allow us to finally test slave news service
in a useful manner. (we've been led to believe that this is already
allocated to us)
o No RS6k AFS server
Replacing the RS/6000 AFS server will allow us to run AFS on the
IS-supported platform, alleviating several maintenance problems (in
particular, currently our PTS database is `scrod' because of this,
and no one (i.e. Richard) seems to be able to help.
o WWW can grow (and grow..)
Adding the new `Meta' server, in addition to providing NFS service,
gives us a place to run the <a href="http://www.mit.edu:8001/stats">
ever-growing</a> World Wide Web server.
