SAS Hard Drives: 15,000 vs. 10,000 RPM

[muk]

Server hard drives are migrating from 3.5" to the faster and more-efficient 2.5" form factor. But what's the performance difference between 15k and 10k RPM in the SAS space?

http://www.tomshardware.com/2007/10/10/sas_hard_drives/index.html

 

[BKD]

I read somebody’s coming out with a PCIe controller that can handle 1200 MB/s sustained transfer speeds. Now try and figure how to saturate that with our CAV hard drives where the more drives you add, the larger the range between mininium and maximum transfer speeds grows. Sure, these SAS drives might get up to 90 MB/s on the high end, but what’s the low end? If you want 1200 MB/s sustained speeds then you have to design the array using that low end value, which sort of sucks. I think it works out to be be like 20 drives for 1200 MB/s vs about 12 drives to reach that on the outside tracks of the disks. So you either settle for 75% of 1200 as the average with 12 drives, or you have to push 8 extra drives and waste all that speed off the top.

But what if we did an opositional RAID 0 where you build the array in pairs that transfer from opposite sides of the disk-inside on one and outside on the other; in this way we get a virtual CLV drive performance where speeds are constant at the median value, between the high and low speeds. I’m using 60 MB/s and 90 MB/s for that, so 150 MB/s(2x75) flat would be the theoretical on an oppositional raid 0 pair. That means you could saturate your 1200 MB/s controller with 16 drives.

That makes more sense to me than an actual CLV drive because CLV drives attain a constant speed by slowing the media down as the heads move to the longer, outside tracks. Slowing the media down? How about pulling your head out of your bum? That’s a little aside to caca roaches over at the blu-ray and hd dvd associations’ golf course. Everyone knows that those 2x transfer speeds are half that of 18x DVD, right? They had to slow the media RPM down by a factor of 15 to get transfers that slow, but think of all the ‘new’ products they can release as they continue to ‘improve’(4x, 6x, 8x, 10x, hybrid CAV). They could roll out ‘new’ products for the next ten years. What I’d like to see is Optical RAID 0 using red lazer, regular DVD drives. I bet if we saw that as an option on Voodoo or botique systems then blu-ray and hd dvd would suddenly realize they can spin their media faster. Maybe the new HD optical format is RAID 0 regular DVD, dual DVD discs for movies that play on any RAID 0 optical RAID with 12 disc arrays on high end systems? Since when did speed stop mattering? When the blu-ray and hd dvd ad money started coming in? Tom? Anyone?

 

[three0duster]

The article was refering to notebooks migrating from 3.5 to 2.5 harddrives. When was the last time a notebook had a 3.5 harddrive in it?

 

[Turas]

I would love to see you pit 4 of those MTRON SSD drives in a raid 0 against both of these. I am currently using the 4 10K ones in my machine and am really debating the switch to SSD but am hoping to find some real benches comparing 4 in a RAID 0 just like the ones you have done here to see real apples to apples comparison.

Any chance of getting ahold of some of those to test with the same parameters as this article.

 

[iampowerslave]

At the conclusion it says that the limit for the controller was 300MB/s and that means each drive transfers 75MB/s and each drive (10kRPM) can transfer 90MB/s all alone

SO HOW CAN YOU COMPARE 10K and 15K drives if BOTH are bottlenecked by the card?

Why didn't you used any Adaptec, LSI or other powerful known brand for SAS instead of a crappy controller?

 

[MrCommunistGen]

The article was refering to notebooks migrating from 3.5 to 2.5 harddrives. When was the last time a notebook had a 3.5 harddrive in it?

The article had several typos. Somewhere else they talk about something being "slower" when it should have been "lower". Also, on some of the charts they didn't translate the word degraded from German.

-mcg

 

[MrCommunistGen]

At the conclusion it says that the limit for the controller was 300MB/s and that means each drive transfers 75MB/s and each drive (10kRPM) can transfer 90MB/s all alone

SO HOW CAN YOU COMPARE 10K and 15K drives if BOTH are bottlenecked by the card?

Why didn't you used any Adaptec, LSI or other powerful known brand for SAS instead of a crappy controller?

They did use a reasonably capable controller.

Intel IOP333 processor running at 800 MHz

-mcg

 

[nanoprobs]

These H.D are SATA connection, is it possible to use these H.D in a notebook? since they are both 2.5".

 

[Turas]

No these are SAS and I am not aware of any notebooks with that interface. You can hook SATA drives to a SAS controller but not the other way around.

 

[Casper42]

To say the use of 15K drives is not worth the cost isnt entirely true.

The HP Proliant DL servers have been using 2.5" SAS now for a while and the DL38x and DL58x servers support 8 SAS drives per machine, with the brand new 580 G5 supporting 16.

What we did at my last company was to standardize the 2 Boot Drives (Mirrored) on the 15K parts and then if that server needed more storage that didnt need to be super high speed, you can create a RAID5 (or RAID6) array with the other 6 empty slots and the 144GB 10K drives.

A SQL Server built recently had 2 15K 72s for the OS and App, 2 15K 72s for the Logs and then 4 10K 144s for the Data. Works really well and only takes 2U for this particular server.

Keep in mind though that larger companies will build larger servers with a SAN back end and then dont care all that much about the local disk. a Pair of 72GB 15Ks is great for the boot volume.

-Casper

 

[MrCommunistGen]

These H.D are SATA connection, is it possible to use these H.D in a notebook? since they are both 2.5".

As someone else mentioned they are SAS not SATA. Also the drives are taller than most notebook drives (12.5mm vs 9.5mm I think) and because of the high spindle speed get far too hot for use in a notebook. Power draw must be off the scale compared to standard notebook drives too. Oh, and it will set you back about $450 for a 73GB 15k Savvio.

-mcg

 

[xrodney]

Last month i bought top line 8+4 port Adaptec SAS controler and 2 hitachi 147GB SAS 15k rpm drives whitch can in raid do 496MB/s peek read transfer and 234MB/s average read transfer.
Normaly that will cost me around 1000$ for controler itself and 500-600$ per drive, but i got it with almost 60% discount so i am now hapy user with enterprise sas solution as sytem drive on my PC
I will post graphs with performance later because right now i am at work and dont have remote acces to my home from here.

 

[xrodney]

some litle test here
http://photos.blackstorm.sk/oc4/

 

[pc007]

What size were the drives tested? If it's in there I missed it.
with a CAV drive, bigger means higher throughput... so a 1TB 7200rpm SATA drive should have a higher throughput than any SAS drive available.
Not enough comparrisons that I saw.

 

[xrodney]

I think i posted it in previous post "2 hitachi 147GB SAS 15k rpm drives in RAID 0", and no way any curent 1TB SATA can outperform even slower SAS drive. Diference in performance and throughput is too big.

 

[pc007]

Data throughput read from a 147GB 15000rpm disk will be equal to a 294GB 7500rpm disk. Sounds like simple math to me. More data will be passing under the head per second on a 1000TB disk than is physically possible on any SAS drive... Assuming the same number of heads.

Do you have any solid numbers to backup your claims?

 

[SomeJoe7777]

Data throughput read from a 147GB 15000rpm disk will be equal to a 294GB 7500rpm disk.

Not even close to being true. These drives do not have the same number of heads or platters, and that is generally the case when comparing smaller drives with larger drives.

Take a look at www.storagereview.com, click on the Performance Database link, and look at the Maximum Transfer Rate comparison. The leader is the 300GB Seagate 15K.5 at 135 MB/sec. The second place drive is the tiny 73GB Seagate Savvio 15K.1 at 108 MB/sec.

 

[pc007]

It seems that the Seagate 1000GB drive electronics are limited to 105MB/s transfer rate... as they are quoted on the seagate site as 105MB/s sustained along with the 500GB & 750GB disks. Where as the Cheetah 300GB version 6 has a sustained rate up to 164MB/s. So it looks like the SAS drives are given electronics that can keep up with the disk, where as the SATA disks are limited.... on purpose I think, as the SATA bus is more than adequate, so it is probably to protect the server storage market.
A 1TB should be able to read data faster given the same physical attributes i.e disk diameter, and independantly positioned heads (AFAIK all drives use one head 'arm')
Has anyone pulled a 15k SAS drive apart? just wondered if they have 2 head 'arms' or just one.

 

[SomeJoe7777]

So it looks like the SAS drives are given electronics that can keep up with the disk, where as the SATA disks are limited.... on purpose I think, as the SATA bus is more than adequate, so it is probably to protect the server storage market.

A 1TB should be able to read data faster given the same physical attributes i.e disk diameter, and independantly positioned heads (AFAIK all drives use one head 'arm')

That's what I just said ... those drives do not have the same physical attributes. 15K RPM drives have 2.5" platters, as opposed to 7200 RPM drives with 3.5" platters. Both the Seagate 1TB drive and the 300GB Seagate 15K.5 have 4 platters/8 heads, but the 1TB drive has a huge amount more physical area to store data due to the larger platters. What this amounts to is that the area-density of the platters is about the same. The 15K.5 outperforms the 1TB drive because the platters are spinning twice as fast.

There is no conspiracy to artificially limit the desktop drives' performance. That's as fast as they go. You can't get them to read any faster without either increasing the area-density (which happens every time they come out with a new series), or increasing the spin speed (makes the drive more expensive and reduces total capacity because you have to decrease the platter diameter).

 

[pc007]

Aha, right you are... I had another look at the seagate site and it does have a picture with much smaller platters on the SAS drives.
Although the 500MB, 750MB, and 1000MB SATA drives are all quoted on the seagate site as being 105MB/s sustained... all the same diameter, just more platters and heads. that has to be regulated somehow, or the numbers are wrong. i'll stick with my conspiricy theory
http://www.seagate.com/docs/pdf/datasheet/disc/ds_barracuda_7200_11.pdf

 

[SomeJoe7777]

Although the 500MB, 750MB, and 1000MB SATA drives are all quoted on the seagate site as being 105MB/s sustained... all the same diameter, just more platters and heads. that has to be regulated somehow, or the numbers are wrong.

Why would that be regulated? 105 MB/sec is simply the result of the area-density of the platter * the spin speed.

The higher capacity drives just have more platters & heads, but you can't read from more than one head at a time. The channel controller chip on the logic board can only process data from one head at a time, so you can't get any speed increase from adding platters/heads.

 

[pc007]

The higher capacity drives just have more platters & heads, but you can't read from more than one head at a time. The channel controller chip on the logic board can only process data from one head at a time, so you can't get any speed increase from adding platters/heads.

Oh... that seems rather inefficient, but explains it. Just think how fast it could be...
I couldn't find any document to confirm what you said... do you have any reference material?

 

[SomeJoe7777]

Oh... that seems rather inefficient, but explains it. Just think how fast it could be...
I couldn't find any document to confirm what you said... do you have any reference material?

You're correct, it's difficult to find anything on this, but I was able to scrounge up one thing:

http://reddit.com/info/5zqbi/comments/c02dd4i

The post about 3 down from the top claims that a competitor to Seagate was developing parallel head read/write technology, and Seagate bought them to squash the development.