xfs
[Top] [All Lists]

Re: data loss with delayed allocaton

To: Fong Vang <sudoyang@xxxxxxxxx>
Subject: Re: data loss with delayed allocaton
From: Steve Lord <lord@xxxxxxx>
Date: Thu, 26 Jan 2006 16:03:58 -0600
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <4f52331f0601261025h9002e41q5df1e15888be7b@xxxxxxxxxxxxxx>
References: <4f52331f0601261025h9002e41q5df1e15888be7b@xxxxxxxxxxxxxx>
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Thunderbird 1.5 (X11/20051201)
Fong Vang wrote:
I just read the following article:
http://madpenguin.org/cms/index.php/?m=show&opt=printable&id=6045

Near the end, this line caught my attention: "the delay means that
while a crash will not destroy FS, it might result in significant loss
of data that has not yet been written to the disc."

I'm just wondering if this delay parameter is tunable and what the default is.


Just the same delay as if the data was written out into already allocated
space. Unless you do a write with O_SYNC or O_DIRECT or use fsync, there
is always a delay between a write call completing and the data hitting the
disk. The only difference with delayed allocation is that the allocate
is done with the flush rather than with the write. The timing difference
on the flush to disk is not going to be measurable.

Steve


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