Re: 2.3.x wish list?

Paul M. Hirsch (pauldoom@webcreate.net)
Fri, 21 May 1999 10:37:55 -0500 (CDT)


On 19 May 1999, Stefan Monnier wrote:

> Stefan "who thinks 4K blocks are too big"

I live and die by 4K blocks. Why? Speed. Here is a bonnie run
showing the performance difference betweek 1k and 4k block sizes:

-------Sequential Output-------- ---Sequential Input-- --Random-- NOTES
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
500 8451 35.2 8264 12.3 3828 9.9 8537 34.8 9334 9.4 101.5 1.5 1024K block, 4096b/i
500 8593 31.2 8921 9.4 4611 8.8 11029 41.4 11700 8.7 129.2 1.1 4096K block, 4096b/i

The runs were done on the same empty 2GB partition.
This was on an Alpha with an IBM 10K rpm UltraSCSI drive, but I get the
same sort of difference when I test other systems.
A better bonnie run doesn't gaurantee better real world performance, of
course, but it is a good indication of overall speed under most
circumstances. Plus, fsck takes alot less time to run on it.

Solaris and Irix (XFS) offer block sizes from 4KB to 8KB, and from 512B
to 64KB respectivly. What would be needed for Linux to support at least
8K blocks? If there is a sweet spot above 4KB for many applications,
it could be a real easy way to squeak out better FS performance. At least
it would us the chance to tune with a wider range of options.

-Paul "Sometimes bigger really IS better" Hirsch

------------------------------------------------------------
Paul M. Hirsch Network Engineer EXi, Corp.
smtp-> phirsch@exicorp.com http-> www.exicorp.com
------------------------------------------------------------

-
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/