I don't know; it was an off-hand comment and I haven't brought
it up before for precisely that reason. I didn't mean to imply
that clustering wasn't working at all... I seemed to remember
benchmarks that showed that 4K and 1K blocks with clustering
had very close I/O nubmers, and so I thought that just in case
some other change in the kernel had started defeating
clustering in some cases, it would be best to make sure that
clustering was working correctly in Leanard's benchmarks before
spending too much time hacking at fragments.
Sorry not to have been clear...
michaelkjohnson