Date: Thu, 23 Mar 1995 20:45:18 +1000 From: Bruce Evans <bde@zeta.org.au> To: davidg@Root.COM, roberto@blaise.ibp.fr Cc: hackers@FreeBSD.org, hasty@star-gate.com, peter@bonkers.taronga.com, terry@cs.weber.edu Subject: Re: Why IDE is bad Message-ID: <199503231045.UAA11233@godzilla.zeta.org.au>
next in thread | raw e-mail | index | archive | help
>> It's the metadata I/O that we're talking about. ...and yes, the process
>> will definately see it if the operation involves lots of file creates. It is
>> inherently unsafe to do metadata I/O asynchronously, which is why the manual
>> page says:
>Yeah, we all know that but ext2fs is doing it for ages now and they
>don't have that much more problem with it... It would be great if
>it worked (for the news spool for example).
Why don't they have much of a problem with it? I can think of the following
possible reasons:
1) with more disk updates, there is more likely to be a failure in the
middle of an update.
a) If the failure is in software and unrelated to the file system
(e.g., a panic for a NULL pointer), then synchronous update
guarantees that the file system (but not the data) can be fixed up
by fsck, but fsck will find problems more often and more data will
be lost. More apparent-problems are bad publicity and lost data
is always annoying.
b) If the failure is in hardware, then synchronous update
doesn't guarantee anything. Metadata may be half written and
you'll be lucky if fsck can't read the bad half. The higher
reliablity of modern drives reduces the disadvantage of
synchronous update here.
c) If the failure is in software and is related to the file system,
then synchronous update doesn't guarantee anything. Metadata
may be scrambled and fsck will normally be able to read it and
become confused :-].
2) ext2fs may be more robust. It's never assumed synchronous updates.
3) ext2fsck may be better than fsck.
4) writing the disk more may wear it out faster.
Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199503231045.UAA11233>