Subject: Re: Why the partitioning should stay the same
To: Mike Hibler <firstname.lastname@example.org>
From: Greg A. Woods <email@example.com>
Date: 01/31/1995 16:35:01
[ On Tue, January 31, 1995 at 13:27:51 (-0500), Mike Hibler wrote: ]
> Subject: Re: Why the partitioning should stay the same
> Here are some random thoughts/observations (based on 4.4-lite so Chris or
> someone can correct me if they no longer apply to NetBSD).
One thing you don't mention is how the algorithms behave with multiple
swap partitions (i.e. raw disk swap partitions).
On SysVr3 (and up?), multiple partitions are used alternately for each
swap operation (and perhaps for each block, though I'm not sure of that
exactly). This results in effectively cutting the wait-for-io on swap
operations, for each additional swap partition, assuming decent disk
drivers and lack of spindle head contention on the swap spindles. Even
with head contention on the swap spindles, multiple swap spindles can
reduce the contention significantly, since *every* swap operation isn't
going to the same disk and thus constantly contending with whatever
other partition(s) is(are) on that disk.
Also, as has been mentioned in one of the NetBSD lists before,
traditional VM in SysV resorts to swapping complete processes after a
certain threshold is reached. If you're kernel is properly tuned in
such cases, deadlocked thrashing never occurs. Rumour has it that
NetBSD has at least the stubs of the code required to do this, but that
it's all currently commented out. Does anyone have any suggestions or
comments about this?
Greg A. Woods
+1 416 443-1734 VE3TCP robohack!woods
Planix, Inc. <firstname.lastname@example.org>; UniForum Canada <email@example.com>