softupdates and caches/tagged queuein (was: Softupdates a mount option?)

From: Matthias Andree <ma_at_dt.e-technik.uni-dortmund.de>
Date: Thu, 27 May 2004 04:51:09 +0200
Brad Knowles <brad.knowles_at_skynet.be> writes:

> At 9:15 PM +0200 2004/05/26, Ivan Voras wrote:
>
>>  This has been really nagging me for a long time: Why aren't softupdates
>>  made a mount option (like 'sync' and 'async')? Do I remember correctly
>>  that it is done so in NetBSD (where it's called softdeps), so it's doable?
>
> 	You need to read the documentation from Kirk.  In short, this
> was intentional -- the goal is that this feature will be turned on
> permanently, at which point there's no sense in having yet another
> useless mount option laying around.  IIRC, the license is written such
> that you are not *allowed* to make this a mount option, for this reason.
>
> 	Kirk is really serious about this.  You should read the Ganger &
> Platt paper, as well as the other documentation written by Kirk on this
> subject.

Speaking of which, does this fine file system (softupdates) know how to
use

- ordered tags (SCSI)

- flush cache commands (ATA) (or other barrier approaches)

to enforce write order where necessary?

I'm wondering a bit for one of my SCSI drive defaults to (control page)
queue algorithm modifier = 1 and (caching page) WCE = 1 which means it
can reorder at will unless someone uses ordered tags or the do-this-now
tag (I forgot the name) - and this discussion is a bit "hot" in the
Linux lists ATM.

-- 
Matthias Andree

Encrypted mail welcome: my GnuPG key ID is 0x052E7D95
Received on Wed May 26 2004 - 17:51:40 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:55 UTC