On Mon, Feb 15, 2010 at 01:48:01AM +0100, Frank Steinmetzger wrote:
> Sorry if I reheat a topic that some already consider closed. I used the
> weekend to experiment on that stuff and need to report my results. Because
> they startle me a little.
>
> I first tried different start sectors around sector 63: 63, 64, 66, 68 etc.
> They showed nearly the same results in speed. So I almost thought that my
> drive, albeit being new and of high capacity, is not affected by this yet.
>
> But then I tested my main media partition, which starts in the middle of the
> disk. I downloaded a portage snapshot and put it into a ramdisk, so reading
> it would not manipulate measurements. I also copied a 1GB file into that
> ramdisk to test consecutive writes.
>
> As a start sector I chose 288816640, which is divisible by 64. The startling
> result: this gave the lowest performance. If the partition starts in one of
> the sectors behind it, performance was always better. I repeated the test
> several times to confirm it. How do you explain this? :-?
>
> The following table shows the ‘real’ value from the output of the time
> command. SS means the aforementioned start sector with SS % 64 == 0.
>
> action SS (1st) SS (2nd) SS+2 SS+4 SS+6 SS+8
> -------------+----------+----------+----------+----------+----------+----------
> untar portage 3m12.517 2m55.916 1m46.663 1m35.341 1m47.829 1m43.677
> rm portage 4m11.109 3m54.950 3m18.820 3m11.378 3m21.804 3m12.433
> cp 1GB file 0m21.383 0m13.558 0m14.920 0m12.813 0m13.407 0m13.681
Instead of guessing using this rather imprecise metric, why not just
look up the serial number of your drive and see what the physical
sector size is? If you don't want to open your box, you can usually
get the information from dmesg.
Only caveat: don't trust the harddrive to report accurate geometry.
This whole issue is due to the harddrives lying about their physical
geometry to be compatible with older versions of Windows. So the
physical sector size listed in dmesg may not be the real one. Which is
why you are advised to look up the model number on the vendor's
website yourself to determine the physical sector size.
W
--
Willie W. Wong ww...@math.princeton.edu
Data aequatione quotcunque fluentes quantitae involvente fluxiones invenire
et vice versa ~~~ I. Newton
