xfs
[Top] [All Lists]

Re: [PATCH 09/10] xfs: xfs_fs_write_inode() can fail to write inodes syn

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: [PATCH 09/10] xfs: xfs_fs_write_inode() can fail to write inodes synchronously V2
From: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Date: Thu, 4 Feb 2010 12:36:14 -0500
Cc: Christoph Hellwig <hch@xxxxxxxxxxxxx>, bpm@xxxxxxx, xfs@xxxxxxxxxxx
In-reply-to: <20100203230235.GB5332@xxxxxxxxxxxxxxxx>
References: <1265153104-29680-1-git-send-email-david@xxxxxxxxxxxxx> <1265153104-29680-10-git-send-email-david@xxxxxxxxxxxxx> <20100203112753.GA19996@xxxxxxxxxxxxx> <20100203205648.GA23116@xxxxxxxxxxxxx> <20100203230235.GB5332@xxxxxxxxxxxxxxxx>
User-agent: Mutt/1.5.19 (2009-01-05)
FYI I did some benchmarking on this, and the syncmodes 2 and 5 of
fs_mark, which use sys_sync regress almost 10% on my test setup
with this patch.  The barriers are only a small part of it,
from instrumentation it seems like the constant log forces don't
really help.  Now given that we only get data integrity writes
from sync_filesystem do we really need to bother with catching
all that pending I/O here?  It would be much easier to rely
on ->sync_fs to do that for us once, which is does anyway.

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