How to Speed up your Hard Disk :-) (Zak's Partition Theory)

[Zak33]

OK: Theory...bust is apart if you so wish

New hard drive looks like this, though the platter count may vary from 1 to 4 (I guess)



Now, let's say 100 people across the world, with XP, all decided to partition their new drives: 1/4 and 3/4. The FIRST partition that they make is 1/4 of the size and the rest is the second. In every case the drive has 2 platters.

How does the Hard drive look? The red sections are the 1/4 of the volume.

---------------------------------------
Does it split the partiton with a WEDGE on both platters, looking like this?


--------------------
or
------------------

Does is partition the drive by using half of one of the platters as a SPLIT?


-------------------
or
-------------------

Does it use the outter RING of ONE PLATTER?


-------------------
or
-------------------
Is it totally RANDOM?


------------------
or, as I think it does....
-----------------

Does is use the OUTER of BOTH?


------------------
(or
does it do it another way?)

The outter edges are not just moving faster, but for a set size will also be NARROWER than further into the platter, meaning the read heads don't need to move so far, and with the extra speed the platters are travelling at, the data speed increase is a given.


QUESTION: Does XP always do it that way? Are the HDD's themselves hardcoded to start at the outter edge? Am I just flukey, or is it a useable tool for everyone?

[Steve]

The internals of a disk's operation will be completely abstracted from the operating system. (edit: that is assuming the OS doesn't care about CHS/LBA addressing all that much)

All the operating system can do is address the drive. How those addresses relate to physical positions on the disk is entirely up to the manufacturer.

But this is how it will work:

Nearby addresses will be physically close to each other. The time to move something versus the time to do something electronically is huge... electronically is so much quicker.

So before moving the read/write heads, it'll swap from one platter head to another.

That makes Zak33's final image more or less the right one.

I'm now going to plug a spare SATA drive I have into a RAID controller and run some IOMeter tests on it.

Back in 20.

[MSIC]

It'll either be the outer ring of one platter, or the very bottom drawing, but i'd be curious to know which (and indeed how or whether a utility such as partition magic could force some particular use).

[Queelis]

was it deliberate?

Can you install the same game on both partitiond and time the boot loads times and the level load times?

Well, yes, it was deliberate, but there are some issues.

First one the hard drive being 12gig, makes it difficult to install games

Second, the partitions' have had their space reallocated, so that may have messed things up.

[iranu]

I find this quite intruiging. I'd love to know. Outter ring of one or both platters would seem the most logical. I'm gonna go googling for the answer.

[aidanjt]

If we assume that hard drive manufacturers know how to make HDDs perform at their best (and I think it's fair that they do), the bottom image would be pretty much how hard drives operate across the board. Clearly it's optimal to do concurrent reads across all the platters at once.

[Steve]

Reading a little more, I expect LBA-48 will be used at the lowest level of the OS. LBA addresses are arranged such that as the address increases you move through the sectors... ie around the disk. Once you've gone around you change head (onto another platter/side) and then over all the sectors again.

Once you've exhausted all that you have to move to a different cylinder... that's the bit that requires physical movement of the heads, hence why you do it less often for sequential reading

So yes, the last image is the 'correct' one.

[Zak33]

If we assume that hard drive manufacturers know how to make HDDs perform at their best (and I think it's fair that they do), the bottom image would be pretty much how hard drives operate across the board. Clearly it's optimal to do concurrent reads across all the platters at once.

ok...well I agree, and I think its prolly right...BUT to take advantage of this, if you follow my basic sumise, and create the FIRST partition as about 1/5 to 1/4 of the drive's capacity, and use that partition for your speed related data, you will probably come outr very well.

QUESTION: Why did I spend all this time doing this?

Because I have been using a Raptor for over a year now, and I love it, but it gets slower after it's about 1/3 full, and I always need another drive for Data, films, music etc. Plus I've gone SFF and dont have rom for 4 or 5 drives. Most people use 320 or 250 gig drives as they are best value at the mo. But to have a fast boot drive is great.

So I wondered: Can a £50 hard disk perform nearly as fast as a Raptor and a spare disk? So I got a 400 gig AAKS WD drive and tried. And to be honest..it's not far short. When the Raptor has a fresh install on it, it's lightning, but it soon slows when a couple of games go in too, plus all the XP updates etc.

[iranu]

I have the answer!!

Zone Effects: Modern hard disks use zoned bit recording to allow more data to be stored on the outer tracks of the hard disk than the inner ones. This directly impacts the media transfer rate of the disk when reading one zone of the disk as opposed to another; see here for details. Hard disks fill their space starting from the outer tracks and working inward. This means that if you split a hard disk into three partitions of equal size, the first partition will have the highest transfer rate, the second will be lower, and the third lower still. Therefore, you can put the more important files on the faster partitions if transfer performance is important to you.

So yes Zak you were right and actually had evidence from nero confirming it.

From Storage review. A site I had bookmarked from over 2 years ago.

[Zak33]

So yes, the last image is the 'correct' one.

Thank you for looking and confirming

The next question is: Can a cheap ass drive perform better by creating this PARTITION on the outter edges, and so force the system to always have it's speed-orientated data in that fast zone?

In fact, without going mad (for which I am known, sadly) on the subect, can a cheap ass £54 drive
http://www.scan.co.uk/Products/Produ...oductID=567858

compete, daily, with a Raptor http://www.scan.co.uk/Products/Produ...oductID=410986 for £95, AND give the end user a spare 300+ gig of space?

Prolly not quite, but it's going to be a close call and it's worth helping people get the most from thier drives.

Wonder if DR can get a clear topped Raptor to watch the head action in?

[Steve]

OK, I've done some quick testing.

I used the IOMeter setup detailed in the second table on this page.

I setup a 20GB partition at the beginning of the disk, then a 190GB partition in the middle, and a 30GB one at the end.

This was on an old Maxtor 250GB SATA drive, 7200rpm with 16MiB cache.

I tested the two end partitions with 1GB test files in IOMeter.

Here are the results:

       First partition     Last partition
Read:       67.23MB/s        44.20MB/s
Write:      67.69MB/s        44.51MB/s
General:    16.86MB/s        13.46MB/s

The "last partition" is a third slower at sustained reading and writing. And we're not even at the end of the disk, assuming that the test file appeared towards the beginning of the 30GB partition.

For "general" performance testing, there is more randomness. That levels the playing field a little, giving a deficit of less than a fifth, because there's more physical movement to access data, slowing both cases down significantly.

So we can say the following things:

• Partitions start towards the outside of the disk and work inwards
• This is supported by what we know about LBA addressing used by disks.
• So we can assume LBA address 0 is at the very edge of the disk, on one of its platters, in one of its sectors.
• Because disks spin at a constant angular velocity, data can be transferred quicker to/from the outer edges, as they will be moving quicker over the read/write heads.
• The above is only true where ZBR (Zoned Bit Recording) is used. This puts more data in the outer sectors of a disk, because outer sectors have a bigger surface area. All disks use ZBR, unless you're using a 5.25" 4MB drive from the 80s.
• For sequential reading/writing, there can be a big performance difference between the outside and inside of a disk.

This isn't ground breaking news, because if you do an HDTach run across a full disk, a lot of the time you'll see it tail off as you get further along the disk.

Bottom line is, if you can keep your more frequently used data towards the beginning of the disk, it should read quicker.

De-fragmenting tries to do this to an extent, but controlling partition size is also a way to force the issue.

However, you should probably let the OS stay at the beginning, because some of the primitive bits of an OS that are used to do the initial boot strapping can't always see the whole disk. (There are workarounds to this, however).

edit: I should also add that if I had a Raptor to test against, it would likely massively own the Maxtor in the "General" test.

[Zak33]

This isn't ground breaking news, because if you do an HDTach run across a full disk, a lot of the time you'll see it tail off as you get further along the disk.

Bottom line is, if you can keep your more frequently used data towards the beginning of the disk, it should read quicker.

De-fragmenting tries to do this to an extent, but controlling partition size is also a way to force the issue.

True, in fact years ago Norton had a Utility for Win98 to move the SwapFile to the outter edge. I'm deliberately creating the partition for both reasons.....rotational speed AND to stop the read heads needing to move laterally, which should ALSO keep the fragmentation down


Thanks for researching bud. Mine is all sumise and trial and error. So thankyou for puting it into context

[aidanjt]

This is nothing new really, UNIX admins have been using partitioning strategies for decades, on optimal linux servers, /boot, swap, and root are regularly the first partitions, followed by /var, then /usr/misc., then /home, for the same reasons outlined here.

[Steve]

So let's hammer out partition strategy, shall we?

How big does your XP partition need to be?

Mine is 50GB but I reckon 25GB would be fine if I was more prudent with my housekeeping. The OS and applications will be their usual responsive self.

Next, have a partition appropriately sized for games. This could be quite big if you have a lot of games... many take up a hell of a lot of space these days.

Then, use the rest for stuff that doesn't need rapid access. Movies, music, backups... none really depend on speed, stick em at the end

And aidanjt is right, *nix users love their partitions.

[Zak33]

So let's hammer out partition strategy, shall we?

How big does your XP partition need to be?

Mine is 50GB but I reckon 25GB would be fine if I was more prudent with my housekeeping. The OS and applications will be their usual responsive self.

Next, have a partition appropriately sized for games. This could be quite big if you have a lot of games... many take up a hell of a lot of space these days.

nope. Not quite. Most people have 2 HDD's not one.

Have XP on the small fast partition of one, but keep it a sensible size so that other apps, such as PaintShop Pro, Office, Skype etc open like lightning. Use the rest of that drive for data, films

Have the second drive partitioned as well, and use the fast one for games, making it big enough to never run out of space. Use the rest for data, films etc.

Get clever and store stuff on one drive that will be acessed by the other drive..ie movies on the large patition and the movie player on the small partition on the other drive

Or am I being anal?

[Steve]

Sounds good, but movie player/movies will make no difference.

Image/movie editor vs. scratch disk will though. As will page file

(now I'm being anal)