From owner-freebsd-current  Fri May 30 03:43:04 1997
Return-Path: <owner-current>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.5/8.8.5) id DAA10445
          for current-outgoing; Fri, 30 May 1997 03:43:04 -0700 (PDT)
Received: from Sisyphos.MI.Uni-Koeln.DE (Sisyphos.MI.Uni-Koeln.DE [134.95.212.10])
          by hub.freebsd.org (8.8.5/8.8.5) with SMTP id DAA10437
          for <current@freebsd.org>; Fri, 30 May 1997 03:42:56 -0700 (PDT)
Received: from x14.mi.uni-koeln.de (annexr2-40.slip.Uni-Koeln.DE) by Sisyphos.MI.Uni-Koeln.DE with SMTP id AA12730
  (5.67b/IDA-1.5 for <current@freebsd.org>); Fri, 30 May 1997 12:42:29 +0200
Received: (from se@localhost) by x14.mi.uni-koeln.de (8.8.5/8.6.9) id MAA03325; Fri, 30 May 1997 12:42:22 +0200 (CEST)
X-Face: "<d]#=8pzx);RzeqSKI86OVa7=!0/(uRa.+B.9Z9\eNUn@UG?!`y7yt2dFNn%k4'.}](uE%
 yCO)$e&Y1%3EO~ifu6Q-#YUM&JZ't,}JkPnAz,8Dj33u%@GBi%[Y#LHz$]h7a<p4)-jKI7~sKjlP-^
 EvA[G;]v&0]W!EL%shs,{7x0|oqN4YVIs5,NI#,V{9"WF):5&RkOhyj*#-IAG}Tnu;YCF,d
Message-Id: <19970530124220.36367@x14.mi.uni-koeln.de>
Date: Fri, 30 May 1997 12:42:20 +0200
From: Stefan Esser <se@FreeBSD.ORG>
To: Poul-Henning Kamp <phk@dk.tfs.com>
Cc: Bruce Evans <bde@zeta.org.au>, current@FreeBSD.ORG
Subject: Re: disk cache challenged by small block sizes
References: <199705291228.WAA11504@godzilla.zeta.org.au> <373.864918382@critter.dk.tfs.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
X-Mailer: Mutt 0.74
In-Reply-To: <373.864918382@critter.dk.tfs.com>; from Poul-Henning Kamp on Thu, May 29, 1997 at 05:06:22PM +0200
Sender: owner-current@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

On May 29, Poul-Henning Kamp <phk@dk.tfs.com> wrote:
> 
> >ufs with a block size of 4K was about the same speed as ext2fs with a block
> >size of 4K.  ufs with a block size of 8K was significantly (25%) slower.
>               ===========================================================
> 
> Uhm, isn't that rather obvious ?

Ahemm ???

> The chances of being able to do one DMA into two (physically) consecutive 
> pages are very very slim as far as I can tell, so you will generally get
> better performance when you do it in page size chunks ?

Sure!

> In other words, drivers should be optimized for two back-to-back 4K
> transactions to different pages, not for 8k transactions to physically
> contiguous pages.

Well, sorry, hmmm, ... When did you write a bus-master 
SCSI driver, last time ?  ;-)

For a "n" page transfer, n+1 scatter/gather table entries 
are provided. This allows for the transfer to start at 
an arbitrary byte offset into the first page, and covers
n-1 full pages, and possibly a final partial one ...

Regards, STefan