RAID Experiment: striping the outsides of platters = speedup?

[Tesselator]

I believe spanning fills one drive and then moves onto the next.

Yup. And zero performance increase. Zero.

Also, you will get the same results in your XBench scores for sequential I/O on the 1st partitions you made as you would if you had not partitioned the drives at all. Depending how random tests are performed in XBench this may be true for the random I/O as well.

What you're doing is called short-stroking and this is best done with the first 8% to 10% of the platter surface as a maximum. So that would be about 75 GB to 90 GB respectively. 150 GB is too big a chunk to really get the max out of a short-stroke unless these are 2 TB or 1.5 TB drives. Of course that's assuming you want the max. Using smaller chunks of 75 GB or less will also impede the remaining partition area less. Likewise with accessing both partitions at the same time for whatever reasons. You want the smallest size you can possibly get away with that's still useful. For example 30 GB would be great for the OS. This is 3-drive RAID0 right? So that's a total of 90 GB. If it's scratch disk then three 10 GB partitions would be about right and would really scream! Etcetera.

It's not correct however to say that short-stroking doesn't affect real-world performance. It does. How much depends on what else is on the rest of the drive(s) and how/when it's accessed in relation to the short-stroked partitions. It may be very much better and it may actually be worse.


.

 

[Loa]

It's not correct however to say that short-stroking doesn't affect real-world performance.

I think that what VR meant was that by having 2 partitions (one fast, the other one slower) you "force" some data to be slower than it could have been. Unless your first partition is 100% filled (but that's never going to happen), you essentially create an unused zone on the fast part of your drive (first partition).

Thus everything you put on your second partition ends up being slower than it could have been without a partition. So overall, by forcing some data to be in a slower part of the drive, you get a slower average speed.

But that being said, I still partitioned my RAID0 in two volumes. That's because the only data I'm going to put on the second (slower) partition is completely performance-blind. In my case, it's essentially downloaded video files and some back-up.

I couldn't care less about drive performance on those files, and even by pushing them out to the slowest part of my 4 disk RAID0, they still read/write twice as fast as they used to on a single drive.

Loa

 

[nanofrog]

I think that what VR meant was that by having 2 partitions (one fast, the other one slower) you "force" some data to be slower than it could have been. Unless your first partition is 100% filled (but that's never going to happen), you essentially create an unused zone on the fast part of your drive (first partition).

Thus everything you put on your second partition ends up being slower than it could have been without a partition. So overall, by forcing some data to be in a slower part of the drive, you get a slower average speed.

But that being said, I still partitioned my RAID0 in two volumes. That's because the only data I'm going to put on the second (slower) partition is completely performance-blind. In my case, it's essentially downloaded video files and some back-up.

I couldn't care less about drive performance on those files, and even by pushing them out to the slowest part of my 4 disk RAID0, they still read/write twice as fast as they used to on a single drive.

Loa

Loa, is the second partition array set backup data also on the first array (separate arrays on the same drives, but contain some of the same data)?

Assuming this is the case, then that's a mistake. That is, if one disk goes, the data on both arrays are shot. Unless the backup data is on independent drive(s), it's not going to help you if you run into a problem that's a dead drive.

 

[Shake 'n' Bake]

With meaningfull power you mean top end for G5 PMs? Where I live you can get a 2,3 GHz DP or DC for 450-700€. We tend to have used prices very simillar to $ prices in the US. So I would be surprised if you could not get a better G5 deal. Actually I guess we will see a lot of panic G5 sales with Snow Leopard and there may be some surprisingly good deals on a privat basis.

I'll have to look at them. Even though they're leak prone, I've always wanted a liquid-cooled model for some reason.

My neighbor works for Apple, so he might be able to grab me a G5 or Mac Pro that is "participating" in the recycling program.

 

[Loa]

Assuming this is the case, then that's a mistake.

You're right! Old habits from when I had 2 drives I guess! It was not my main back-up (which is on a dedicated drive), but I've cleared it up after reading your post.

Thanks

Loa

 

[VirtualRain]

I did some more reading on short-stroking. It seams that it really benefits Windows installs because Windows will distribute files all across the volume it's provided for the C: drive. For example, the page file is setup 1/2 way into the volume. Thus short stroking really benefits Windows OS volumes as it contains all the files to a relatively small surface area on the disk, reducing the seek times.

The key question for us is how does OSX behavior compare to Windows? Does it deposit files all over the volume space it's given or does it fill a volume from outside to in? If it behaves like Windows, then it would benefit from short-stroking as well. If it's more organized and fills a volume from outside to inside then it won't matter whether the system partition is 10% of the disk or the whole disk, performance will be the same.

 

[gugucom]

Dp

 

[gugucom]

I allways asked myself why designed the Bootcamp/EFI to install Windows allways in the lowest partition. This would make a conveniant explanation.

It would also stipulate that short stroked Windows disks are faster in MBR than in GUID partition table.

 

[DoFoT9]

REALLY interesting stuff learning all about the fastest positions on the drive, will be implementing a two 500GB drive, two partition RAID0 for my hackintosh (OSX on outer, XP on inner tracks). great reading, thanks everyone for the info

 

[Tesselator]

REALLY interesting stuff learning all about the fastest positions on the drive, will be implementing a two 500GB drive, two partition RAID0 for my hackintosh (OSX on outer, XP on inner tracks). great reading, thanks everyone for the info

You know of course that you don't have to partition them to get that speed tho right? I mean the inside 20% is always the same speed - partitioned or not. If you never fill your drive(s) more than 20% you'll always get that speed. Thus more reasons to get bigger drives even tho you don't intend to utilize all that space. For example if it's two or three 2TB drives then the inner 20% is like 400MB X Number of drives. Just never use more than that and you'll be fine - no partitioning needed.

 

[DoFoT9]

You know of course that you don't have to partition them to get that speed tho right? I mean the inside 20% is always the same speed - partitioned or not. If you never fill your drive(s) more than 20% you'll always get that speed. Thus more reasons to get bigger drives even tho you don't intend to utilize all that space. For example if it's two or three 2TB drives then the inner 20% is like 400MB X Number of drives. Just never use more than that and you'll be fine - no partitioning needed.

Yup, I did know that! it's just that i wish to implement it for my hackintosh, and i need a dual boot system, OSX + XP - so partitioning it evenly just seems like a good idea.

400MB? woohoo

up a few posts ago you said 8%-10 for fast access, but now you say 20%. how much of a performance decrease would one see if they used the first 20% to "short-stroke"?

 

[nanofrog]

Yup, I did know that! it's just that i wish to implement it for my hackintosh, and i need a dual boot system, OSX + XP - so partitioning it evenly just seems like a good idea.

400MB? woohoo

up a few posts ago you said 8%-10 for fast access, but now you say 20%. how much of a performance decrease would one see if they used the first 20% to "short-stroke"?

XP or Win7? Which is it?

I doubt the 10% to 20% would make much difference, perhaps not even notable (depends on disk size, as Tesselator restricted the comment to a 2TB disk). It would need to be tested on a case by case basis, as I'd expect it to become noticable as the drive size becomes smaller. So less of an issue on a 2TB disk (multiplatter) vs. say a 250GB model (single platter units). The physical location on the platter(s) would play into it, even if partitioned.

 

[DoFoT9]

XP or Win7? Which is it?

haha Win7, Win7!!!!! i keep getting confused there are so many!

I doubt the 10% to 20% would make much difference, perhaps not even notable (depends on disk size, as Tesselator restricted the comment to a 2TB disk). It would need to be tested on a case by case basis, as I'd expect it to become noticable as the drive size becomes smaller. So less of an issue on a 2TB disk (multiplatter) vs. say a 250GB model (single platter units). The physical location on the platter(s) would play into it, even if partitioned.

aaahh i understand. the 500GB drives that i am using has two platters (i am fairly certain). so the slow down would be not as significant as on a single platter 250GB drive, correct?

 

[nanofrog]

aaahh i understand. the 500GB drives that i am using has two platters (i am fairly certain). so the slow down would be not as significant as on a single platter 250GB drive, correct?

If the single platter disk uses the same platter density, then Yes (also assume the same spindle speeds, and servo motors,...are the same).

It's when you're comparing different densities that makes it a bit more complicated. Say a 4 platter drive @ 250GB per platter (1TB) vs. a 3 platter 1TB (334GB per platter) vs. a 2 platter 1TB (500GB/platter). The last one would be the fastest in the comparison, and work backwards to the next and so on. If the spindle speeds and other mechanicals are different, it gets even more complicated. So testing is the best way to know hard numbers, but assuming you can keep as much as possible at the same spec, more platters of the highest density would yield faster results for random access when partitioning in this manner.

Hope this makes sense.

 

[DoFoT9]

If the single platter disk uses the same platter density, then Yes (also assume the same spindle speeds, and servo motors,...are the same).

i see, ill make sure i purchase the same drive.

It's when you're comparing different densities that makes it a bit more complicated. Say a 4 platter drive @ 250GB per platter (1TB) vs. a 3 platter 1TB (334GB per platter) vs. a 2 platter 1TB (500GB/platter). The last one would be the fastest in the comparison, and work backwards to the next and so on. If the spindle speeds and other mechanicals are different, it gets even more complicated. So testing is the best way to know hard numbers, but assuming you can keep as much as possible at the same spec, more platters of the highest density would yield faster results for random access when partitioning in this manner.

Hope this makes sense.

are there any single 500GB platter drives haha? that would be nice and fast

it makes perfect sense, as the technology increases it allows more bits to be stored on the same size - physically smaller bits = faster read/write times as there is less movement etcetc.

 

[nanofrog]

i see, ill make sure i purchase the same drive.

That's why it's best to use identical drives in an array.

It's not always possible, particularly for expansion and the existing drives are no longer available, or so hard to find, the price is too high. But good planning can help prevent mixed sets.

are there any single 500GB platter drives haha? that would be nice and fast

it makes perfect sense, as the technology increases it allows more bits to be stored on the same size - physically smaller bits = faster read/write times as there is less movement etcetc.

Yes. The Seagate 7200.12 500GB is a single platter. I'd have to look a little deeper into WD's offerings, as I think the 2TB models are, but not sure if the lower capacity drives have switched yet.

 

[DoFoT9]

That's why it's best to use identical drives in an array.

It's not always possible, particularly for expansion and the existing drives are no longer available, or so hard to find, the price is too high. But good planning can help prevent mixed sets.

well your point seems to have suddently become valid, because the local company i buy from only sells the 7200.12 model!!! bugger..

is there any risk in combining these??

or should i combine a WD + Seagate (i am fairly certain both are 2x250GB platters)

Yes. The Seagate 7200.12 500GB is a single platter. I'd have to look a little deeper into WD's offerings, as I think the 2TB models are, but not sure if the lower capacity drives have switched yet.

not sure about WD either. they are my preferred brand atm though (until i have trouble with them).

 

[nanofrog]

well your point seems to have suddently become valid, because the local company i buy from only sells the 7200.12 model!!! bugger..

is there any risk in combining these??

or should i combine a WD + Seagate (i am fairly certain both are 2x250GB platters)

It depends, as the slowest disk will essentially set the speed for the set. Now if you continue to replace old drives for multiples of the new model (online expansion, if the array type supports it), you'd eventually get all drives as the same model.

If both drives are similar in specs, such as RPM (spindle speed), seek times (servo), platter count and capacity (both translate to platter density), you shouldn't have any real issues as they'd be close enough that one of them wouldn't present as a bottleneck to the other drive(s) the set.

not sure about WD either. they are my preferred brand atm though (until i have trouble with them).

I'm sticking with WD for SATA drives as well. RE3 line for RAID, Blacks for backup/misc, and a lonely Velociraptor meant as a boot drive for a non RAID'ed OS (OS X is intended for it).

 

[VirtualRain]

I'm sticking with WD for SATA drives as well. RE3 line for RAID, Blacks for backup/misc, and a lonely Velociraptor meant as a boot drive for a non RAID'ed OS (OS X is intended for it).

Are the RE3's the same drive as a Black with a different TLER setting, warranty, and price?

 

[DoFoT9]

It depends, as the slowest disk will essentially set the speed for the set. Now if you continue to replace old drives for multiples of the new model (online expansion, if the array type supports it), you'd eventually get all drives as the same model.

perfect answer, thanks. i better just use my two existing drives then, and use the new drive as in place of the other one (if that makes sense haha)

If both drives are similar in specs, such as RPM (spindle speed), seek times (servo), platter count and capacity (both translate to platter density), you shouldn't have any real issues as they'd be close enough that one of them wouldn't present as a bottleneck to the other drive(s) the set.

they are identical AFAIK - i will check the version numbers once i get home to clarify that they have the same platter sizes, if they are different i will put off doing it until i can afford purchasing the same sized ones.

I'm sticking with WD for SATA drives as well. RE3 line for RAID, Blacks for backup/misc, and a lonely Velociraptor meant as a boot drive for a non RAID'ed OS (OS X is intended for it).

they just seem more reliable currently. i am quite fond of Hitachi, but i have been reading reports about their drives failing more often and their service department being poor.

 

[Tesselator]

Yup, I did know that! it's just that i wish to implement it for my hackintosh, and i need a dual boot system, OSX + XP - so partitioning it evenly just seems like a good idea.

400MB? woohoo

up a few posts ago you said 8%-10 for fast access, but now you say 20%. how much of a performance decrease would one see if they used the first 20% to "short-stroke"?

Hehehe I meant GB not MB.

And no, I'm not saying 20% is the best or fastest region. It's just an example number I picked out of the air.

<10% is the best.

 

[DoFoT9]

Hehehe I meant GB not MB.

i find myself doing it all the time

And no, I'm not saying 20% is the best or fastest region. It's just an example number I picked out of the air.

Ahh ok fair enough, you can never tell over text-based communication if you're "umming and arring" about these things. but it's somewhere around those numbers, i guess only a physical test will tell!!

<10% is the best.

oh really? sure you didn't pick that out of the air 10% would be too small for me and hardly useful at all, at a glance anyway.

 

[Tesselator]

oh really? sure you didn't pick that out of the air 10% would be too small for me and hardly useful at all, at a glance anyway.

Yes, I'm sure. People who short-stroke for speed stay well below the 10% mark. It's pretty common information. It's obvious when benchmarking the drive zones too. But that can be pretty big these days. Assuming you went the full 10% on a 4 drive array of 1TB drives that's about 350 GB.

It's about 650 GB on a 4 x 2TB drive array.

The first 10% (or a little less actually) of a drive is 3 or 4 times faster than the next 70%. The second band of 10% is only about 1.2 to 1.5 times faster than the next 60% of the drive. The second to last 10% zone is 1.2 to 1.5 times slower than the previous 60% band and the last 10% band is 2 or 3 times slower than the previous 70% zone.


.

 

[DoFoT9]

Yes, I'm sure. People who short-stroke for speed stay well below the 10% mark. It's pretty common information. It's obvious when benchmarking the drive zones too. But that can be pretty big these days. Assuming you went the full 10% on a 4 drive array of 1TB drives that's about 350 GB.

have you translated that into actual GigaBytes? (e.g. 4 x 1TB HDs in an array = 3.5GB total storage, 10% of that is 350GB.) is that correct?

for the quote below then, should it not be 700GB space? (e.g. 8TB total minus 1TB, = 7TB of usable data, %10 = 700GB) or am i missing something . sorry just wanting to make sure i understand it fully.

It's about 650 GB on a 4 x 2TB drive array.

point proven, that is fairly large when used in a 4x 2TB drive array, not that i can afford that. i am currently limited to 2x 500GB HDDs, so that number diminishes quite significantly, to the tune of 93GB or something according to my calculations - that would be ok if i work on one video project/audio project at a time (which is what i would most likely do).

 

[Tesselator]

Sure, whatever.

But 500GB drives are like, soooo three years ago. I've seen people on here recently buying less that 1TB drives but I can't for the life of me understand it - at all. It's just dumb. To me HDDs come in 3 sizes ONLY. 1TB, 1.5TB, and 2TB. Nothing else is even a consideration. 500GB... even for free that's too lame. IMO anyway.