xfs
[Top] [All Lists]

Re: [PATCH 2/2] XFS: Mark no-more-exist xfsbufd related sysctl parameter

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: [PATCH 2/2] XFS: Mark no-more-exist xfsbufd related sysctl parameters as obsolete.
From: Satoru Takeuchi <takeuchi_satoru@xxxxxxxxxxxxxx>
Date: Mon, 10 Dec 2012 10:58:08 +0900
Cc: Christoph Hellwig <hch@xxxxxxxxxxxxx>, xfs ML <xfs@xxxxxxxxxxx>
In-reply-to: <20121210013048.GK15784@dastard>
References: <50C521D7.5060202@xxxxxxxxxxxxxx> <20121210013048.GK15784@dastard>
User-agent: Mozilla/5.0 (Windows NT 6.0; rv:15.0) Gecko/20120907 Thunderbird/15.0.1
(2012/12/10 10:30), Dave Chinner wrote:
> On Mon, Dec 10, 2012 at 08:42:15AM +0900, Satoru Takeuchi wrote:
>> From: Satoru Takeuchi <takeuchi_satoru@xxxxxxxxxxxxxx>
>>
>>
>> xfsbufd is removed since commit 43ff2122e6492bcc88b065c433453dce88223b30.
>> But there stille be two sysctl parameters about it
>> (fs.xfs.{xfsbufd_centisecs,age_buffer_centisecs}.)
>>
>> This patch marks these parameters as obsolete. In adition, prints
>> a warning message if someone tries to use these parameters.
> 
> While we currently don't use them, I'm not sure they are obsolete.
> We need to integrate the AIL flushing with time-based log idling
> (i.e.  flushing to an empty state), and part of that work is making
> the AIL log items have a age-based flush component.
> i.e. how often it will wake to try to flush/idle the log when there
> is no pressure, and how long items should wait before being flushed
> even when there is activity.
> 
> Those two parameters are exactly what the xfsbufd parameters have
> historically controlled, so rather than introduce new sysctls for
> the same purpose, I was planning on just re-using them. Hence I
> suspect that marking them obsolete is probably the wrong thing to do
> right now - adding the time-based component to the AIL flushing
> would be a better solution.

OK, got it.

Satoru


<Prev in Thread] Current Thread [Next in Thread>