[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Disk geommetry, was Re: Kernel Upgrade: Why?



On Thu, 22 Apr 1999, Ookhoi wrote:

> Well, there was a discussion here about a benchmark Linux vs NT, and
> some people here said that the preformance of Linux could have been
> affected by the fact that Linux was near the center, and NT on the outer
> side. 

Probably not the reason. 

> And the data on the outer side passes the heads much faster than the
> data on the inner side. But then, there is much more data on the outer
> side, and a piece of data on the outer side will go round in the same 
> amount of time as a piece of data on the inner side..

I am not aware of any disks that use a higher density recording format for
the outer tracks than they do for the inner tracks.  As far as I am aware
(and I really haven't paid much attention to such things since ST-277's
were state of the art) the bit density of the outer tracks is LOWER than
the bit density of the inner tracks.  That's because the outer tracks are
physically larger, but they hold the same number of bits.

Not that it matters.  The whole disk spins as a single unit so even if
there were more bits on the outer tracks, you'll still wait the same
amount of time (on average) for the sector you want to come around.  Read
on, and I'll explain.
 
> So, is there an advantage if whe put for example swap at the outer side
> of a disk?

NO!

Look, the access times for disk are dominated by two times, the time to
seek to the correct track and the time to wait for the data to come around
again on the disk.  The time it takes the data to come around on the disk
is, on average, one half of the time it takes for the disk to go around
once.  That's independant of everything else and is a fairly short time,
anyway.

The time it takes to seek to the correct track depends upon where you're
seeking from and where you're seeking to.  Obviously, if the heads happen
to be at innermost cylinder, it will take longer to seek to the outermost
cylinder than if the heads were in the middle or toward the outside.  So,
for higher performance in a situation where you're too cheap to add enough
RAM, you'll want the swap file near where the heads are likely to be.

You can also turn that around.  Seeking to the middle from either extreme
is likely to be faster than seeking to the other extreme.  (This works for
both average and worst-case times.)

Predicting where the heads are likely to be takes some doing, especially
on systems with effective disk caches, but you can take some educated
guesses.  The middle of a disk is a better guess than either extreme, but
isn't necessarily the best guess.  If you spend a lot of time reading and
writing (especially writing) files from a particular partition, you might
want to put the swap file near that partition on the principle that the
heads are likely to be near there anyway, so it should reduce the amount
of time waiting for any given swap.

An extreme example of this would be where you dedicate an entire drive to
a (fairly small) swap partition.  That's how the news servers I use do it.
For something less extreme, I kind of like the recommendation made by OS/2
gurus:  Their advice was to put the swap file in the most used partition
on the least used drive.  You might try something like that where you put
the swap partition in the middle of a disk that isn't used for very much.

In short, my recommendation for boosting the performance of a computer
that uses a significant amount of swap is to add RAM to the computer.

HTH.  HAND.
-- 
Jonathan Guthrie (jguthrie@brokersys.com)
Brokersys  +281-895-8101   http://www.brokersys.com/
12703 Veterans Memorial #106, Houston, TX  77014, USA


Reply to: