Received:  by oss.sgi.com id <S554259AbRASSZB>;
	Fri, 19 Jan 2001 10:25:01 -0800
Received: from sgi.SGI.COM ([192.48.153.1]:56348 "EHLO sgi.com")
	by oss.sgi.com with ESMTP id <S554254AbRASSYs>;
	Fri, 19 Jan 2001 10:24:48 -0800
Received: from madurai.engr.sgi.com ([163.154.5.75]) 
	by sgi.com (980327.SGI.8.8.8-aspam/980304.SGI-aspam:
       SGI does not authorize the use of its proprietary
       systems or networks for unsolicited or bulk email
       from the Internet.) 
	via ESMTP id KAA03842
	for <linux-xfs@oss.sgi.com>; Fri, 19 Jan 2001 10:24:35 -0800 (PST)
	mail_from (ananth@sgi.com)
Received: from sgi.com (mango.engr.sgi.com [163.154.5.76]) by madurai.engr.sgi.com (980427.SGI.8.8.8/970903.SGI.AUTOCF) via ESMTP id KAA57736; Fri, 19 Jan 2001 10:18:16 -0800 (PST)
Message-ID: <3A688567.9CC4CE7A@sgi.com>
Date:   Fri, 19 Jan 2001 10:20:23 -0800
From:   Rajagopal Ananthanarayanan <ananth@sgi.com>
X-Mailer: Mozilla 4.72 [en] (X11; U; Linux 2.2.16-4SGI_20smp i686)
X-Accept-Language: en
MIME-Version: 1.0
To:     Christoph Hellwig <hch@caldera.de>
CC:     Marcelo Tosatti <marcelo@conectiva.com.br>, linux-xfs@oss.sgi.com
Subject: Re: pagebuf page cleaner and page aging
References: <200101191510.f0JFAHs02250@jen.americas.sgi.com> <Pine.LNX.4.21.0101191123270.5416-100000@freak.distro.conectiva> <20010119183803.A23862@caldera.de>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-linux-xfs@oss.sgi.com
Precedence: bulk
Return-Path: <owner-linux-xfs@oss.sgi.com>
X-Orcpt: rfc822;linux-xfs-outgoing


There's been several mail exchanges already on this thread.
Let me try to summarize:

1. The page cleaner should walk the dirty list of pages.

2. There are 2 reasons to start write-out earlier than on
   just memory pressure/aging information alone. (a) to minimize
   data lost due to delalloc pages not written to disk (b) to minimize
   write pressure on the disk .. you don't want to fill-up a 1GB system
   with delalloc pages & starting to write out at 20 MB/sec to disk.

3. Its almost tempting to make kswapd do the delalloc conversions,
   since it walks the inactive dirty list through page_launder().
   However, if you think about it, the conversion is an operation
   that logically sits between make a page delalloc & writing it out to
   disk. IMO, we need a seperate daemon to perform the conversions.
 
4. Another issue is proper write-throttling to avoid situations as in 2 (b).
   Perhaps we can make balance_dirty() to synchronously perform some conversions?

   
-- 
--------------------------------------------------------------------------
Rajagopal Ananthanarayanan ("ananth")
Member Technical Staff, SGI.
--------------------------------------------------------------------------