ext3fs slowness -- how things proceed

[Matthias Andree]

After Greg reported the problem to the ext3-users mailing list, Andrew
Morton picked up the discussion and can see parts of the problem -- with
some 2.5. version though which IMHO behaves differently.

I've also posted the simbf test code to ext3-users and my own findings
which essentially group file system/hard drive interfaces into two
categories:

"quick" with ext2 + ide, reiserfs + ide, ext3 + scsi
"slow"  with ext3 + ide

The interesting thing is "vmstat 1" output simultaneous to the bfsim
code, it shows data trickle to hard disk in 5 s intervals (around 20 or
30 blocks/s, may also be induced by other software on my machine) for
configurations in the "quick" category, and constant writes of around
600 blocks/s for ext3+ide.

Strange. I hope this data gets Andrew Morton on the right track.

There be dragons.

-- 
Matthias Andree