From mboxrd@z Thu Jan  1 00:00:00 1970
Message-ID: <63626928d1b9d7d7a5001593df460222@coraid.com>
To: 9fans@cse.psu.edu
Subject: Re: Re: Re: Re: [9fans] system crash during compile
From: Brantley Coile <brantley@coraid.com>
Date: Thu, 13 Jul 2006 09:44:42 -0400
In-Reply-To: <ee9e417a0607122109k2004f53cj1cb0a1d30d79c220@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset="US-ASCII"
Content-Transfer-Encoding: 7bit
Topicbox-Message-UUID: 7e313564-ead1-11e9-9d60-3106f5b1d025

> Swap has never worked.  It mostly works, but there have
> always been hard-to-reproduce and then even harder-to-debug
> little issues, and I just don't trust it at all.  With memory so
> cheap, there is little motivation to fix it.

That explains our mail processing problems that we had a few years ago
when the incoming load was so high that it exceeded local store and we
turned on swap, but things were still funky.  (wound up limiting the number
of connections allowed to port 25 at a given time.)

Might I suggest we remove swap from the system, at least remove it
enough so people don't fall prey to it?

I would argue backing store is a thing of the past and that Sandy
Fraser's objections to the Plan 9 team early on, as I understand his
objections to be, were correct.  Let's just use paging to manage the
store in the box and to perhaps demand load text.  We're a long way
from tiny, expensive local stores and slow drums.

Let's just take it out.