Re: Data authentication for geli(8) committed to HEAD.

From: Peter Jeremy <peterjeremy_at_optushome.com.au>
Date: Fri, 9 Jun 2006 19:59:23 +1000
On Fri, 2006-Jun-09 12:15:33 +0930, Daniel O'Connor wrote:
>I think that in reality with a modern high capacity disk you don't get atomic 
>writes at all because they all re-write whole tracks.

AFAIK, this applies to _all_ modern disks, whether they're SATA, PATA or SCSI.

>Yes this violates the assumption soft updates makes, I believe the only way 
>around it is to buy SCSI drives (not because they're SCSI per se, but because 
>they're smaller capacity so they don't do this)

Not totally.  Soft updates requires that the device driver inform the
filesystem when a write has been committed to non-volatile media.  The
device driver can only do this if the underlying device reports this.
All SCSI devices do but few ATA devices support tagged queueing so the
device driver can only tell that the data has been written to the
disk cache, not that the data has been committed.

You can work around this (at the cost of reducing write performance by
about an order of magnitude) by disabling write caching.

-- 
Peter Jeremy

Received on Fri Jun 09 2006 - 07:59:58 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:57 UTC