Simon Burge wrote:
> MLH wrote:
> 
> 
>>One thing I noticed with recent kernels is that ata disk performance
>>has decreased markedly on one of my machines, supposedly the fastest
>>one of three fairly equivalent machines.
>>[ ... ]
>>    -------Sequential Output-------- ---Sequential Input-- --Random--
>>    -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
>>100 45140 59.8 55066 42.7 53620 50.8 73079 84.2 359877 92.3 12769.8 65.3
>>[ ... ]
>>100 25066 22.6 23989  7.3 29172 10.4 95010 96.2 415142 85.9 11538.2 45.3
>>[ ... ]
>>NetBSD 1.6ZG (APMLM) #0: Wed Jan  7 17:25:13 CST 2004
>>total memory = 511 MB
>>avail memory = 493 MB
> 
> 
> Note that you're testing a 100MB file on a box that can effectively have
> (a little bit less than) 493MB of buffers.  Look at the seeks-per-second
> column of your bonnie results - no (non-solid-state) disk can do 10,000
> seeks a second!

If the disk is not involved I think the question about the slowdown
is even more interesting.  Something bad has happened in the kernel.

	-- Lennart