----- Original Message -----
From: "Dimitry Sibiryakov" <SD@...>
To: <ib-support@yahoogroups.com>
Sent: Wednesday, February 20, 2002 9:15 AM
Subject: RE: [ib-support] File system


> On 19 Feb 2002 at 11:16, Darryl VanDorp wrote:
>
> > My preference would be a journaling filesystem
> > over the ones you mentioned and ext2 over
> > NTFS just 'cause i like linux better.
>
> Once upon a time Lilya Kozlenko said that NT4 with NTFS
> can rollback a couple of filesystem transactions after a
> hard failure (reset, trap and so on). If any file was
> created or extended just before the crash, these changes
> would disappear and DB file could be unrecoverable damaged.

It seems that you do not understand how journalling FS work. Only
inconsistent chages in metadata only are rolled back to move file system to
previous stable state. NTFS does not journal changes in in file data because
it would cost too much and therefore it doesn't guarantee no data loss so
one should not expect this without hardware redundance and UPS.
In the case you described FAT would get much more damage since file catalog
entries would no longer reflect actual file size and any file operation
would further destroy the file (or even other files).

> So, FAT could be a more reliable choice. Besides, FAT is a

No, NTFS is certainly a choice for production systems over FAT.

> bit faster because of the lack of overhead to journaling.

No. NTFS performance might be slightly better or worse depending on file
size and other factors. In general, the difference is negletable.

> She also said that a raw device is the best choice.

Is there any testing data that prove significant (more that few percent)
performance boost on raw devices for IB?

Regards,
Aleksey Karyakin