questions regarding Journalling-FSes and w-cache reordering

questions regarding Journalling-FSes and w-cache reordering

[Oliver S.]

I've got a question regarding the behaviour of w-caches with ATA-harddisks:
Is a ATA-harddisk allowed to reorder the operations from in w-cache ? On one
side, this would be an opportunity to optimize head-movements (although a good
cyclical-scan drive-scheduler would gain the same if it can rely on a linear
sector-mapping), but on the other side, this would compromise the two-phase
commiting of journalling FSes and render it useless. So I doubt that reorder-
ing is allowed with ATA-HDs. Are my assumptions right ?

RE: questions regarding Journalling-FSes and w-cache reordering

[Mudama, Eric]

> -----Original Message-----
>
> Is a ATA-harddisk allowed to reorder the operations from in 
> w-cache ?

ATA hard drives are allowed to reorder/merge/etc their write caches if write
cache is enabled.  With write caching enabled, there is no guarantee that
dirty data will be flushed in any specific order, nor does the ATA protocol
support any such ordering beyond the global flush cache command.

No order preference is supported in TCQ either except for the optional FUA
extensions for ATA-7.

There is nothing in the ATA specification that guarantees data integrity at
any given instant in time on the media without a flush cache.  (e.g. I am in
the process of writing data A, and data B is an arriving overlapping write
splice, what do I actually put on the media and in what order)

Power cycling a dirty, busy drive, especially in the middle of error
recovery/reallocation, is always a dangerous proposition.


--eric

--
Eric Mudama                              Maxtor Corporation
eric_mudama@maxtor.com      2452 Clover Basin Drive - BC163
303.682.4758 (x4758)                     Longmont, CO 80503
                      cell: 303.478.3180
 

RE: questions regarding Journalling-FSes and w-cache reordering

[Alan Cox]

On Mer, 2003-04-09 at 20:52, Mudama, Eric wrote:
> ATA hard drives are allowed to reorder/merge/etc their write caches if write
> cache is enabled.  With write caching enabled, there is no guarantee that
> dirty data will be flushed in any specific order, nor does the ATA protocol
> support any such ordering beyond the global flush cache command.

To cheer people up further by the way, not all ATA drives support cache
flush, some drives dont support write cache disable and others implement
write/verify as write.

So basically you can't do it.

RE: questions regarding Journalling-FSes and w-cache reordering

[Torrey Hoffman]

On Wed, 2003-04-09 at 13:12, Alan Cox wrote:
> On Mer, 2003-04-09 at 20:52, Mudama, Eric wrote:
> > ATA hard drives are allowed to reorder/merge/etc their write caches if write
> > cache is enabled.  With write caching enabled, there is no guarantee that
> > dirty data will be flushed in any specific order, nor does the ATA protocol
> > support any such ordering beyond the global flush cache command.
> 
> To cheer people up further by the way, not all ATA drives support cache
> flush, some drives dont support write cache disable and others implement
> write/verify as write.

Can anyone recommend some decent 120 GB or larger IDE drives which don't
play these games?   

I'd like to build a reliable ~1 TB RAID 5 from IDE drives.  I will
gladly give up performance for high reliability and low cost.

My current 240 GB RAID 5 is built from five 60 GB Maxtor IDEs, with a
Reiser FS on it.  I use hdparm to disable write cache... unless Maxtor
ignores that?  Does anyone know?   I can check the exact model numbers
of the drives if it matters.

Is there documentation anywhere that says which drives cut corners?
Any other suggestions on how to build a cheap but reliable RAID?

Thanks in advance,

-- 
Torrey Hoffman <thoffman@arnor.net>

RE: questions regarding Journalling-FSes and w-cache reordering

[Alan Cox]

On Mer, 2003-04-09 at 22:35, Torrey Hoffman wrote:
> Can anyone recommend some decent 120 GB or larger IDE drives which don't
> play these games?   

On the newer drives you should be fine. You may need to tell the drive to
remember settings over a reset.

Alan