Re: imapd and synchronous writes

John Gardiner Myers (jgm+@cmu.edu)
Fri, 15 Mar 1996 13:30:53 -0500 (EST)


Alan Cox <alan@cymru.net> writes:
> Not true. If you have any modern disk controller that re-orders writes
> from its internal cache then you potentially lose.

My impression was that other filesystems waited until the disk
controller reported back that it had committed the writes to
non-volatile storage before returning. It didn't matter how the
writes were ordered.

> There are file systems
> other than the Linux one that do async writes and will have the same
> properties.

Which ones?

> Add it to gnu autoconf

I would if I knew how. What is it that am I supposed to test for?

"Theodore Ts'o" <tytso@MIT.EDU> writes:
> Well, now that you know, you can do it. An application which does an
> open() on the directory, followed by a fsync(), shouldn't break on a
> broad range of unix systems.

I need a "feature" test I can check for in order to enable this code
only on the systems that need it. I'm not particularly enthusiastic
about taking a performance hit on my primary platforms (Solaris, etc)
in order to pander to a fringe OS that decided to play fast and loose
with the filesystem semantics.

-- 
_.John G. Myers		Internet: jgm+@CMU.EDU
			LoseNet:  ...!seismo!ihnp4!wiscvm.wisc.edu!give!up