Re: HD performance question
On 19 Nov 2002, Olaf Meeuwissen wrote:
> Daniel Pittman <daniel@rimspace.net> writes:
>> On Mon, 18 Nov 2002, Olaf Meeuwissen wrote:
>> > I'm a minimalist and rolled my own kernel. It was absolutely bare
>> > bones and that had a noticable effect on hard disk performance. So
>> > I tinkered around a bit with kernel options and tested performance
>> > with hdparm -tT. Now I'd like to know what all those numbers mean
>> > and if they are reasonable (for my Dynabook SS S4/275PNHW).
>> >
>> > I've repeated all tests five times and dropped outliers. With my
>> > initial kernel I get
>> >
>> > ~110 MB/sec for buffer-cache reads
>> > ~ 2 MB/sec for buffered disk reads
>> >
>> > After tinkering I get
>> >
>> > ~ 55 MB/sec for buffer-cache reads
>> > ~ 14 MB/sec for buffered disk reads
>> >
>> > Question 1: Which of the two is "better" and why?
>>
>> The second, because 2MB/second is PIO, while the second is DMA
>> transfers. So, buffer-cache reads may be slower... but not using 100%
>> CPU when you touch the hard disk is worth it. :)
>>
>> > Question 2: Can I do better than this?
>>
>> No.
>
> Because?
Because your drive is clearly running in an DMA mode, to give that speed
from the drive buffer, which means that anything else you do will have
little or no effect on the overall performance of the disk.
[...]
>> One of the best ways to improve laptop performance is more memory
>> because, always, the hard drive performance is not going to be great.
>
> Maxed that one out already.
*nod* About the only thing which will improve performance, then, is to
go for the idea of a multi-drive RAID setup, as suggested elsewhere, or
to buy an external Firewire or USB 2 hard drive enclosure, put a desktop
drive into it, and use that. :)
Daniel
--
It's my guess that those cutting-edge artists who attack tradition secretly
believe tradition will survive to enshrine them as the wild and crazy geniuses
who destroyed it.
-- Brad Holland
Reply to: