SUMMARY: Swap Space Philosophy

From: Cary Talbot <talbot_at_hl.wes.army.mil>
Date: Mon, 20 Nov 1995 10:14:56 -0600 (CST)

Little did I know that such a seemingly simple question would generate
so many responses!

This list is a great resource

Original question:

Are there any benefits or potentially any penalties for spreading swap
space across two different SCSI controllers (ie one internal disk with
a swap partition and one external disk on a different SCSI controller
with a swap partition)?

ANSWER:

Yes, in most cases it is ok and even desirable (read: FASTER IO) to spread
swap space across two SCSI controllers.

Here's some snipped replies:

>From "Paul E. Rockwell" <rockwell_at_rch.dec.com>:

Merely spreading the swap across disks on the same bus does not
allow simultaneous writing of data. The SCSI bus can only be used for one
transfer at a time. You may lessen rotational or seek delays by spreading
the swap on the same controller.

However, splitting the swap to disks on different SCSI controllers will
allow swapping I/O's to run in parallel.

>From Jim Belonis <belonis_at_dirac.phys.washington.edu>:

There is always some advantage to spreading I/O load across as many devices
and controllers as possible.
But typical SCSI controllers can easily handle 2-3 disk drives with essentially
no degredation, so I wouldn't spend extra money for another SCSI controller
unless you have 3 or more very active disks already.

If you already have a 2nd SCSI controller, by all means put your extra
swap on it.

>From Allan Small (Cafeine propelled) <small_at_gidday.ENET.dec.com>:

This all depends on how much swapping (or other IO) you are performing. Keep
in mind that you can saturate a SCSI bus with about 3-4 disks. If you have
more than 3 heavily used disks on a bus, and you expect to see some activity
with the swap space, then it may be worthwile moving it to another bus.

and From Ray Stell <stellr_at_smyrna.cc.vt.edu>:

>all this is in the system performance and tuning guide, AA-Q0R3D-TE,
>but the rules discussed in the following old post still stand:
>
>Summary of best configuration for swap space:
>=============================================
>In short...
>
>1. 1.2 bad, 1.3 good
>2. allocate to multiple drives when possible
>3. allocate across multiple busses when possible
>4. try to avoid other i/o moving the heads of the drives
> used for swap
>In long...

(he then gives detailed explainations of the above points from
various users and DEC folks - if you want a copy, email me direct and
I'll happily send you these valuable bits...)

Many thanks to all who replied:
"Paul E. Rockwell" <rockwell_at_rch.dec.com>
Blue Moon Network Administrator <root_at_net.bluemoon.net>
"Anthony D'Atri" <aad_at_nwnet.net>
nick_at_alldata.com (Frank "Nick" Riley)
alan_at_nabeth.cxo.dec.com
rwa_at_cs.athabascau.ca (Ross Alexander)
chu_at_musp0.Jpl.Nasa.Gov (Eugene Chu)
Tom Holt <T.Holt_at_uea.ac.uk>
Jim Belonis <belonis_at_dirac.phys.washington.edu>
Allan Small (Cafeine propelled) <small_at_gidday.ENET.dec.com>
Ray Stell <stellr_at_smyrna.cc.vt.edu>
Menelaos Karamichalis <mnk_at_wuerl.wustl.edu>

Cheers,
Cary
_____________________________________________________________________
Cary A. Talbot, Hydraulic Engineer US Army Corps of Engineers
ATTN: CEWES-HH-G Waterways Experiment Station
3909 Halls Ferry Road
Vicksburg, Mississippi 39180
phone:(601)634-4286 fax:(601)634-4208 email:talbot_at_hl.wes.army.mil
Received on Mon Nov 20 1995 - 18:07:05 NZDT

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:46 NZDT