SUMMARY: 6 disks, 2 SCSIs, how to config swap?

From: J. Porter Clark (Porter.Clark@msfc.nasa.gov)
Date: Sat Sep 03 1994 - 02:32:57 CDT


A few days ago, I asked this question:

> My 600 MP model 512 now has 128 MB RAM, two 1.3 GB disks on the CPU
> board's SCSI, and four 1.05 GB disks on a Fast SCSI. I'm running
> Solaris 2.3.
>
> This machine is being used as a news server.
>
> I know about how much swap space I want, but how should I spread it
> out? A small partition on each of 6 disks? (Seems excessive,
> somehow.) Two partitions, one partition per SCSI? I know I don't want
> multiple swap partitions on the same disk.

I think I received every possible answer! Here's a summary:

* Split swap among all 6 disks
* Put only enough swap on the root disk (with the slow SCSI) to boot
* Put swap on disks that are being accessed less frequently
* Going beyond 2 swap partitions will make no noticeable improvement
* More than 2 swap partitions per SCSI gains nothing
* More than 1 swap partition per SCSI gains nothing
* Half each only on the 2 CPU SCSI's, other SCSI will cause extraneous
contention
* Put swap on the fast disks only

As you can see, these answers are somewhat inconsistent.

I also noted the following comments:

* Need > 6 GB for a news server? [No, most of it is reserved for a
less frequently used application.]
* Swap is split on 1 MB boundaries
* 128 MB news server won't do much swapping
* Not as critical with Solaris 2

ddl@placeri.wimsey.com (Duncan Laidlaw (Vcr Computers)) posted a
similar, though by no means identical, question not long ago, and he
sent me the responses he had received.

In addition, Kevin Weinrich <kbw2@helios.ath.epa.gov> sent me the
following tip. He is using SunOS 4.1.3, but I think that the basic
idea should work on Solaris 2.3.

> There is a rarely-used but sorely-missed-if-unavailable advantage
> to having a swap partition that you can get along without in a pinch,
> if it's the first partition on the disk (even if it's called "b").
> I had this on a system and was able to re-boot without using that
> swap partition, re-partition the disk to create 3 little partitions
> a, b, and d on that disk, newfs a&d, dump my existing root and /var
> partition to that disk, install the boot block, and re-boot from
> that disk in order to re-format my original boot disk. If this is
> done carefully, you don't *need* a tape backup or a CD-ROM to boot
> miniroot off of. Neat, huh?

Here's what I ended up doing. Based on the divergence of opinion in
the responses I received, it's possible that I will get more inputs
saying that what I have done is somewhere between non-optimum and
extremely stupid. I can see that some experimentation might yield a
more definitive answer. But what I did seems to work fine, and this
machine is being used too heavily for me to do much experimentation.
Those of you who would like to discuss alternative approaches might
want to take it to comp.sys.sun.admin.

What I did was to put approximately equal-sized partitions on three
disks: one on the 1.3 GB boot drive and two on the 1.05 GB drives.
The idea was to try to make the data rate as high as possible without
quite saturating the SCSI buses. I'm not at all sure that this is the
right way to look at it, because the data rate when doing small random
transfers is low compared to the peak rate. Therefore, it might very
well be that many small partitions would not end up saturating the SCSI
buses when used as swap.

The following people responded:

szh@zcon.com (Syed Zaeem Hosain)
jerry@soul.ampex.com (Jerry Stachowski)
Mike Raffety <mike_raffety@il.us.swissbank.com>
ddl@placer1.wimsey.com (Duncan Laidlaw (Vcr Computers))
Kevin Martinez <lps@rahul.net>
Dan Stromberg - OAC-DCS <strombrg@bingy.acs.uci.edu>
iain@shiraz.jcorp.dialix.oz.au (Iain Massey)
paulo@nova.dcc.unicamp.br (Paulo Licio de Geus)
Eckhard.Rueggeberg@ts.go.dlr.de (Eckhard Rueggeberg)
Ron Hall <thorn@cc.mcgill.ca>
rwolf@dciem.dnd.ca (Robert Wolf)
Marco Pineda <mcp4@columbia.edu>
John DiMarco <jdd@db.toronto.edu>
Kevin Weinrich <kbw2@helios.ath.epa.gov>

-- 
J. Porter Clark       jpc@hammer.msfc.nasa.gov or porter.clark@msfc.nasa.gov
NASA/MSFC/EB33        (205)544-3661
Huntsville, AL 35812  FAX: (205)544-9582



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:09:09 CDT