> On Mon, 28 Jan 2002, Linus Torvalds wrote:
>> (Also, I'd like to understand why some people report so much better
>> times on dbench, and some people reports so much _worse_ times with
>> dbench. Admittedly dbench is a horrible benchmark, but still.. Is it
>> just the elevator breakage, or is it rmap itself?)
On Mon, Jan 28, 2002 at 04:37:02PM -0200, Rik van Riel wrote:
> We're still looking into this. William Irwin is running a
> nice script to see if the settings in /proc/sys/vm/bdflush
> have an observable influence on dbench.
They have observable effects, but the preliminary results (the
runs are so time-intensive the repetitions needed to average out
dbench's wild fluctuations are going to take a while) seem to
give me three intuitions:
(1) the bdflush logic is brittle and/or not sufficiently adaptive
(2) dbench results fluctuate so wildly it's difficult to
reproduce results accurately
(3) dbench results fluctuate so wildly it obscures the true
performance curve
The winner of the first round seems to be 7 0 0 0 500 3000 28 0 0
Cheers,
Bill
