Cached file read performance with 6.2-PRERELEASE

[Mark Kirkwood]

JoaoBR wrote:
> On Wednesday 20 December 2006 07:37, Mark Kirkwood wrote:
>> JoaoBR wrote:
>>> On Tuesday 19 December 2006 22:05, Mark Kirkwood wrote:
>>>> $ dd of=/dev/null if=/tmp/file bs=32k                  # read it
>>>> 819200000 bytes transferred in 1.801944 secs (454620117 bytes/sec)
>>> hum, look my releng_6:
>>>
>>> # dd of=/dev/null if=/c/c1/file bs=32k
>>> 819200000 bytes transferred in 0.896507 secs (913768635 bytes/sec)
>> Hmm - so your machine reads memory twice as fast as mine, which is
>> great! but I'm not sure it actually shows anything useful ... let me
>> guess - a P4 with DDR or DDR2 memory??? maybe if we look a bit harder at
>> buffer cache performance you could get 1.2GB/s or more - wouldn't that
>> be a good thing?
> 
> 
> oh! if it is useful or not seems to be your call, you was the one who said 
> linux is twice as fast as freebsd :)

Oh dear :-), I'm actually a FreeBSD guy, I just happen to need a Linux 
box because the product I'm working on does not build on FreeBSD 
(yet....heh heh).

> 
> it's a dual-opteron with ddr400 I just compiled world with no tweaks and 
> standard newfs options
> 
> if you're interested I send new results after completing my setup
> 

Sorry JoaoBR - I am interested!

I was however trying to point out that as your machine is different from 
mine (opteron and ddr*400* as opposed to PIII and pc133), the fact that 
it is faster is not telling us anything about whether releng_6 
performance on cached file reads could be improved!

In fact if you note that the PIII HW *can* actually do 700MB/s, it 
suggests that your HW is capable of considerably more than 900MB/s - 
given that opteron's have excellent cpu to memory bandwidth, and the 
speed of your memory!

Cheers

Mark