xfs
[Top] [All Lists]

Re: Follow up -- Re: Files on XFS not safe?!

To: Dan Hollis <goemon@xxxxxxxxx>
Subject: Re: Follow up -- Re: Files on XFS not safe?!
From: Seth Mos <knuffie@xxxxxxxxx>
Date: Fri, 07 Dec 2001 00:07:25 +0100
Cc: Xianglong Yuan <yuanx@xxxxxxx>, <linux-xfs@xxxxxxxxxxx>
In-reply-to: <Pine.LNX.4.30.0112061310500.30527-100000@anime.net>
References: <4.3.2.7.2.20011206084345.02e5c618@pop.xs4all.nl>
Sender: owner-linux-xfs@xxxxxxxxxxx
At 13:11 6-12-2001 -0800, Dan Hollis wrote:
On Thu, 6 Dec 2001, Seth Mos wrote:
> Tux2 seems to be a really good idea for static content systems but I am
> afraid that databases will not like phase trees.

Why would they care?

Databases work synchronously most of the time for the data but they might use a buffer for storing intermediate before they are actually written out. This happens with transactions. they store all the data which needs to be committed in the end to actual database or be discarded when it doesn't procede.


I can't see that it would make a single difference at all.

It probably won't, I think my brain just switched on (the horror).

I was mostly thinking of it as something that writes in batches which should be faster but the actual latency might get higher. But I might have interpreted the scheme too course. If it is something that runs every 1/20 second you would get a continous stream of IO but I am still wondering about when multiple file being touched at once (or multiple databases if you like) and meaning that a read must wait before the tree has been commited to disk which might take long if you toch a lot of files at once, before returning the read call.

These are a lot of "mights" and still havn't read the article. Include standard idiot disclaimer here.

Cheers

--
Seth
Every program has two purposes one for which
it was written and another for which it wasn't
I use the last kind.


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