Subject: Re: soft updates Re: Summer of code ideas
To: None <email@example.com>
From: Johan A.van Zanten <firstname.lastname@example.org>
Date: 04/07/2007 16:15:56
Dieter <email@example.com> wrote:
> So no one believes Usenix papers by respected authors, but we have one
> report from an unknown user with unknown hardware that softdep allegedly
> caused lossage and therefore no one trusts softdeps?
I've never had any dataloss problems with softdep mounted FFS file
systems, but up until very recently, i've never had a file system bigger
than 20 GB or so.
I hadn't considered how significantly drive size has increased since
softdeps was written 8 years ago, and perhaps hard drive performance, too.
Drive size matters because people may be storing larger files (like
digitized music, video, etc.) and therefore if 20 seconds of metadata is
lost, and multiple GB files are referenced in that lost metadata, then the
loss could be 10s or even 100s of GB.
In some ways, drive performance has also improved. I'm seeing
substantially better performance with new SATA drives at a price/GB point
i was never able to personally afford before. If a larger volume of data
is moving through the FFS-softdep code more quickly because of
improvements in drive throughput, then it seems possible that 20 seconds
of metadata now references more file data than it was 8 years ago, when
McKusick presented his paper to USENIX 99, though i'm just specualting --
i haven't looked at the code.