On Tue 08 Jun 1999, Erik E. Fair wrote:

> 	bigger blocks, lower overhead,

Uhm, that was my first thought too on this matter. However, when I started
benchmarking with a kernel with Koji Imada's patches, the result was exactly
opposite... The suggestion was that the drive's firmware might have been
optimized for 512 byte/blocks. I must say that this was an old-drive, maybe
modern drives to better. But I think the above statement should be taken
with caution...

Here was my original Bonnie benchmark - I had my mail archives handy ;-)

             -------Sequential Output-------- ---Sequential Input-- --Random--
             -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine   MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
100-1K   100   384 60.0   380 11.6   290 13.6   509 99.3   994 23.7  30.0  9.0
                                                                              
100-512  100   583 99.0  1110 63.3   559 47.3   508 99.3  1005 46.6  31.5  8.8
                                                                              
100-512O 100   583 99.1  1114 64.4   558 50.1   508 99.4  1005 46.6  31.6  8.9
                                                                               
Legenda:
    100-1K  : Use patched kernel on a 1K sectored disk
    100-512 : Use patched kernel on a 512B sectored disk
    100-512O: 512B sectored disk, kernel not patched


Leo.