Re: [Exim] Large queue sizes - big directories - and Reiser …

This is a multi-part message in MIME format.
--
[ Picked text/plain from multipart/alternative ]
Actually there are two different elements I'm referring to here; small
file advantages - and large directory advantages of the Reiser file syste.

As to small files - Reise has something called suballocation allowing
small files to share the same block. It ends up reducing total storage
by 5-10%.

Where the real advantage is in handling large directories. If you have
1,000,000 file in a directory, and you try to open the "last one" then
under ext2 you would have to compare 1,000,000 file names before you can
open the file. This is where Exim runs into a wall when queues grow to
be huge.

With Reiser - since it is built on a b-tree database when you go to open
the file - it's almost instant. And that's the real advantage as it
applies to Exim. Also - Reiser never runs out of inodes.


Tabor J. Wells wrote:

>On Thu, Dec 04, 2003 at 08:11:43AM -0800,
>Marc Perkel <marc@???> is thought to have said:
>
>
>
>>Where Reiser really shines is on large directories with small files.
>>Like what happens with big mail queues where some spammer has filled you
>>up with thousands of bounce messages to non-existent email addresses.
>>Reise can be a hundred times faster in those situations.
>>
>>
>
>This is one of the things I've been wondering about for a while. What's the
>definition of "small files" here? 1k? 5k? 25k? If it's say 5k or less where
>it shines and you send a bunch of mail that is 20k each (and by bunch I'm
>thinking like 100-200k messages on a list exploder), does performance tank?
>
>
>
--