Hi Simon, et al,
I used stock kernel.org sources with SGI's XFS patches:
for XFS 1.1, I used the 2.4.18 kernel
for XFS 1.2, I used the 2.4.19 kernel
for XFS 1.3pre, I used the 2.4.21 kernel
The only reason I did this was to be complete with my tests. As I said
in a previous email, I just naively thought that sync would actually
flush all the data to the disk. The disk is an IDE disk with a 2MB
cache.
I thought that the disk cache might be the cause of the results, so
I added sleeps of from 1 - 10 seconds. No difference in my results were
seen.
One of the reasons I included the script was for anyone interested in
the the phenomenon I think I found, to actually run the script on their
system and if different results were obtained, we could figure out what
was different.
Thanks so far to all who've taken up this thread. I like XFS for it's
throughput but if data integrity is not guaranteed through a sync call,
I have my reservations about it.
-blair
|