Re: patch: reiserfs for 2.3.49

From: Chris Mason (mason@suse.com)
Date: Sat Mar 11 2000 - 17:31:54 EST


On Sat, 11 Mar 2000, Chris Evans wrote:
> On Sat, 11 Mar 2000, Chris Mason wrote:
>
> > Hello everyone,
> >
> > ReiserFS for 2.3.49 is now available for testing, comments, and review.
> > You can get the code from:
> >
> > ftp://ftp.devlinux.com/pub/namesys/2.3-beta/linux-2.3.49-reiserfs-3.6.1-patch.gz
>
> Chris,
>
> I'm going to grab this and test this when I get the chance.
>
> Thanks for the detailled post on the status. One piece of information
> was missing though, which I think might be of interest to readers of this
> list; how fast is this beast ;-)
>

Well, it depends. For the most part, it is as fast, or faster than our
2.2 code was, especially for files without tails (larger than 16k). Files
will tails will be slower in the following case:

while() {
   open(file with tail) ;
   lseek(anywhere in the tail);
   write(data) ;
   close(file) ;
}

To write to the tail of the file, we convert it into an unformatted node
(insert indirect item, balance, cut direct item, balance), and then let
the generic file write function do its thing. On file close, the data is
repacked into the tail (insert direct item, balance, cut indirect item,
balance).

This does slow things down by a few percent, but the operations are mostly
done in cache, and it is not as bad as it sounds. There are two possible
solutions: allocation on flush (don't allocation allocate unformatted node
or insert indirect item until things need flushing), or use the address
space operations to write directly into the file tail.

Note, I'm concentrating too much on the downsides at the moment, on
purpose. In general, it is working, and it is fast. We have some
benchmarks on the way.

I should also note the patch I posted has our utils in it (mkreiserfs and
such). We are writing a script to split them out of the patch, and make a
tar file with just the kernel files required to build the utils.

-chris

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Wed Mar 15 2000 - 21:00:20 EST