Re: filesystem overhead on large files

[Brian Tinsley <btinsley@emageon.com>]

We use ReiserFS to store and serve medical images, which can be anywhere 
from 30KB to hundreds of MB. ReiserFS provided the most consistent and 
most often the best performance versuses other viable Linux filesystems 
(we tested ext2, ext3, and XFS - JFS was not mature enough at the time). 
I can also attest to the fact that an Oracle database is *extremely* 
fast when stored on a ReiserFS formatted filesystem.
 

Radim Kolar wrote:

>tests on 100 mb file on ide disk
>hdd: 6306048 sectors (3229 MB) w/81KiB Cache, CHS=6256/16/63, (U)DMA
>
>dt was using async io in this test
> ./dt of=/dev/hdd1 limit=100m bs=16k dispose=keep passes=10 errors=50
> enable=aio
>
>kernel 2.4.19
>=============
>
>Raw disk IO:  /dev/hdd1 
>       read  2260674 bytes/sec
>       write 3542486 bytes/sec
>
>reiserfs:
>       read  2265559 bytes/sec
>       write 3524625 bytes/sec
>ext2:  
>       read  2226276 bytes/sec
>       write 3214847 bytes/sec
>ext3:      
>       read  2189094 bytes/sec
>       write 3201759 bytes/sec
>minix:
>       read  2208303 bytes/sec
>       write 3046710 bytes/sec
>vfat:
>       read  2159045 bytes/sec
>       write 2573193 bytes/sec
>
>i do not have XFS there XFS claims to be filesystem with highest possible
>transfer rate on large files, but this test shows that reiserfs has
>almost zero overhead on large files, so it can be also multimedia fs.
>
>We are considering to use Oracle in raw disk mode (without filesystem)
>and this test shows that you will not get any benefit from that - and
>backup/restore procedure is more complicated.
>

-- 
Brian Tinsley
Chief Systems Engineer
Emageon
http://www.emageon.com/