Subject: Re: Adding B_WRITETHRU -- file system question
To: Alfred Perlstein <bright@mu.org>
From: Manuel Bouyer <bouyer@antioche.eu.org>
List: tech-kern
Date: 12/04/2001 21:01:38
On Tue, Dec 04, 2001 at 01:55:50PM -0600, Alfred Perlstein wrote:
> Ohhhhh...
> 
> we tried that under FreeBSD, the 4.3 release was so ungodly slow
> on IDE disks because of it that we got a LOT of negative PR.

Yes, I already tried that, and it's why I didn't turn in off :)
If we have a B_WRITETHRU, maybe we could issue a FLUSH CACHE when
appropriate. This needs to be tested.

> 
> Linux keeps it on, FreeBSD keeps it on, I think you guys should
> consider leaving it on just to avoid the pain for the users.

If FLUSH CACHE isn't workable I think I'm going to make it configurable.
I've several machine where write speed isn't an issue, but having it fail fsck
because the disk did a bad thing it.

> 
> Although it's less correct to keep write caching on, it's probably
> better, a consumer purchasing IDE equipment usually is going
> commodity and want's the best bang for his/her buck no matter
> the consequences.
> 
> <possible FUD>
> One last point, I seem to remeber our IDE guru lamenting the fact
> that even though he turned writecache off, some "smart" drives
> would in fact turn it back on if there was extremely heavy IO
> pressure on the drive.
> </possible FUD>

OK, thanks for the hint.

--
Manuel Bouyer <bouyer@antioche.eu.org>
--