Re: ext3 File System Performance Issue

Hi Robert,

"Robert Redelmeier" <redelm@xxxxxxxxxxxxxxx> wrote:

Sounds a lot like a newspool. I can't imagine anyone
want a journal on that flood!

Let me put it this way: I would like to build my own swap, so that
it becomes larger in case the swap partition space is not enough.

Does mounting the FS with `noatime` improve things?"

It seems the "noatime" can help improve read performance but writes
are still the same. (removing files should be similar to writes, right?)

I've just done a little experiment: I intentionally generate 500K files
in the /tmp directory and then try to remove them *randomly*.
I found that the removal is extremely slow, until the number of files
drops below 20K. So I believe this is the "threshold" number.

However, when the files are removed in ascending/descending order
of their creation, the removal speed becomes much faster. Even a
"flip-flop" order of file removal (i.e, if files are created from 0, 1, 2,
...., 499999, I remove the files using the order 0, 250000, 1, 250001,
....) works better than random.

I believe this must relate to the data structure used by the ext3, but
I'm not sure the exact reason why.


-- Robert


Best Regards,
LaBird (Benny).
[Email: Remove all numerals to get the correct address.]


.