Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!samsung!uunet!xstor!iverson
From: iverson@xstor.UUCP (Tim Iverson)
Newsgroups: comp.periphs.scsi
Subject: Read-ahead and power failure
Message-ID: <201@xstor.UUCP>
Date: 10 Sep 90 18:23:38 GMT
References: <1990Aug31.211751.10553@simasd.uucp> <2344@sud509.ed.ray.com>
Reply-To: iverson@xstor.UUCP (Tim Iverson)
Distribution: na
Organization: Storage Dimensions, Inc.
Lines: 23

In article <2344@sud509.ed.ray.com> heiser@tdw201.ed.ray.com writes:
>I was told by a Unix consultant that 'read ahead cache' should NOT
>be enabled on Unix systems.  This has something to do with the fact
>that Unix does its own caching (buffering), and if the drive does
>caching too, things could get confused in case of a power failure.

>Bill Heiser
>heiser@tdw201.ed.ray.com
>{decuac,necntc,uunet}!rayssd!tdw201!heiser

This is not true.  Read-ahead caching is purely write-through, so there are
no problems with cache concurrency.  The power fail problems all result from
unix's block device buffering scheme, which is write-back.  Depending on
how your system is used, turning read-ahead on can cause a performance
*decrease* in rare cases due to the increased overhead involved in
performing the read-ahead (e.g. when your system never wants the data the
data that has been cached).  The thing to do is try it and see.

BTW, I'd axe the consultant - paying for misinformation is a bad idea when
you can get it for free from the net :-).

- Tim Iverson
  uunet!xstor!iverson