Re: Large Partition Performance Test

[Oleg Drokin <green@namesys.com>]

Hello!

On Fri, Apr 12, 2002 at 09:21:22AM -0400, Chris Mason wrote:
> > Even withoyt skip_busy mode it will work much better.
> > The problem is on each file creation we start to look for a free block almost
> > from the beginning of disk. using no_unhashed_relocation mount option,
> > we make initial block allocation to be more random and cause less overhead
> > (verified).
> > And in this test we create a lot of 15k files.
> Ok, I'll buy that.  The only way to know for sure is to give it a try or

OF course you'll buy that, because my suggested workaround worked ;)

> profile the run, but I think it is very likely to be allocator related.

First I decided I need a profile run, but once I realised how small created
files are, and looked on the on-disk block distribution chart.
Then I remembered on what basis we choose 1st block of a file to be allocated,
and I realised I do not need profile run ;)

In fact I am interested in similar task performance of ext3, since I've yet
to try it.

Bye,
    Oleg