backtop

Print 22 comment(s) - last by mindless1.. on Feb 14 at 1:58 PM

Seagate's Savvio 10k.4
Customers clamoring for capacity

Solid state drives might be all the rage right now, but traditional magnetic-based storage still rules the roost. Many corporations are moving to a tiered storage approach utilizing SSDs, but they still require massive capacities at a relatively economical price.

Most server farms have moved to the 2.5-inch hard disk drive format in order to increase storage density and decrease power consumption. These smaller drives have less capacity, but you can fit many more of them into a blade server.

Seagate is one of the biggest suppliers of 2.5-inch HDDs spinning at 10k RPMs. The company is introducing its latest Savvio 10k.4 drive today at 450GB and 600GB capacities. It uses 3 platters to achieve this, which is a little strange considering that it is already using a higher areal density in some of its mobile offerings. Competitor Western Digital is also planning to launch a 600GB version of its next-generation Velociraptor 10k drive within the next few months.

The reason is that its customers have been clamoring for more capacity, according to a well-placed source within Seagate. The company is working on another enterprise drive that will use 300GB platters, but that won't be available until later this year.

Instead of waiting, the company went ahead with the technology it had. Enterprise level drives require extensive testing and verification before acceptance, so Seagate was able to collect a lot of data on the new drive.

The Savvio 10k.4 manages to achieve a 2 million hour Mean Time Before Failure and a 0.44% annualized failure rate, making it the first to do so. This gives it 20% higher reliability than any other drive, according to the company. It also has a very low Unrecoverable Error Rate of 1 sector in 10E16.

The drive will be available with either a 4Gbps Fiber Channel or 6Gbps SAS interface, along with a five-year warranty.

Comments     Threshold


This article is over a month old, voting and posting comments is disabled
2.5" Size
By jdietz on 2/9/2010 1:36:53 PM , Rating: 2
I think enterprises use these because you can fit more drives in the same space versus 3.5" drives notwithstanding the higher capacity of those drives. It won't be long now before your home desktop case has a slot for a 2.5" drive.

I think that should read "low error rate" rather than high.




RE: 2.5" Size
By BruceLeet on 2/9/2010 2:09:40 PM , Rating: 2
My Gateway has 6 2.5" slots.


RE: 2.5" Size
By amanojaku on 2/9/2010 2:28:14 PM , Rating: 1
2.5" drives are used in the server space because of the shrinking need for internal storage, and the lower power usage, noise and heat generated compared to 3.5" drives. 2.5" drives provide greater performance, as well; the full stroke seek time is lower (2-7ms compared to 3-11ms) due to the smaller platter size, while full rotational latency is the same at the same RPMs (6ms @10K or 4ms @15K).

Servers are typically SAN or NAS attached, so the servers just need boot and application volumes. Drives in SANs and NASes are 3.5" still; a stack of 3.5" drives provide more capacity than a stack of 2.5" drives in the same space. This is because 3.5" drives provide more platter surface and less "packaging". I agree, however, that in the future 2.5" drives will become standard in desktops, as well, once capacities increase. A big selling point of PCs is the huge drive.
quote:
It uses 3 platters to achieve this, which is a little strange considering that it is already using a higher areal density in some of its mobile offerings.
I'm not sure the author understands the reason for more platters compared to greater density. More density improves capacity greatly while improving performance slightly. More platters, however, reduces the read and write times essentially by doubling, tripling, or quadrupling the number of simultaneous read or write operations. I say essentially, because no disk drive is capable of reading or writing more than one OS I/O at a time; instead, the OS I/O is split at the drive level. For example, a two-platter drive can write 1MiB in four chunks, one 256KiB chunk for each platter side. A four-platter drive would write eight 128KiB chunks simultaneously. Therefore, the four-platter drive has double the sequential read and write performance of the two-platter drive. Random access performance is trickier to measure.


RE: 2.5" Size
By Iketh on 2/9/2010 8:13:34 PM , Rating: 2
You're implying that there is internal RAID in mechanical harddrives. Why in the world do all modern mechanical harddrives have near identical read/write performance, no matter how many platters?


RE: 2.5" Size
By Agripa on 2/9/2010 9:19:45 PM , Rating: 2
quote:
More platters, however, reduces the read and write times essentially by doubling, tripling, or quadrupling the number of simultaneous read or write operations. I say essentially, because no disk drive is capable of reading or writing more than one OS I/O at a time; instead, the OS I/O is split at the drive level. For example, a two-platter drive can write 1MiB in four chunks, one 256KiB chunk for each platter side. A four-platter drive would write eight 128KiB chunks simultaneously. Therefore, the four-platter drive has double the sequential read and write performance of the two-platter drive.


Head alignment issues prevent drives from reading or writing from more than one head at a time. Hard drive spindle and circular runout have been greater than the track spacing for at least 10 years. That is one of the reasons they switched to an embedded servo servo signal from a dedicated servo.

If you want to verify this, just check the sequential read and write speeds for drives in the same series with a different number of heads or platters. They will have almost identical speeds.


RE: 2.5" Size
By tygrus on 2/10/2010 1:16:44 AM , Rating: 2
*** Cannot double STR by doubling heads & platters with current designs.

They can only align one head at a time to use one platter at a time. This means they can only do one IO to one platter even if the drive has 8 heads. If a large read of 8MiB it could read 512KiB from the same track# on each surface with each head and then track-to-track latency to move head to next track# to continue sequential read. Having the same data in less number of tracks and towards the outside of the disk means higher average performance but doesn't change the maximum.

Some time the number of tracks increase more than Kilobytes per physical track when the capacity is doubled between generation.

Track layouts are offset so by the time taken to move track-to-track, the next logical sector is ready to read just after. Any further and average rotational delay and longer seek times affect access times (ie. performance, random reads cannot reach max STR).


RE: 2.5" Size
By mindless1 on 2/10/2010 2:50:46 PM , Rating: 3
1) Server drives are not used because of lower noise. On the contrary servers are built with quite noisy fans often.

2) Full stroke seek time is not different enough to be significant if comparing same RPM drives, as the larger ones have higher data transfer rate and very small reads are cached.

3) Servers are not "typically" SAN or NAS attached. That would completely negate the implied benefit you claimed of performance from it's own HDD.

4) In the future (and now) 2.5" will be a standard for SSD, not mechanical HDDs in desktops. Maybe if you were talking about an extremely long time waiting, many years from now they would be standard in 2.5" mechanical but at that point it is likely the standard will be SSD not mechanical HDD. You conceded it yourself, the selling point of mechanical is large size and 3.5" will always have more size (storage capacity).

5) I'm not sure you understand the reason for more platters either. They use more platters to keep the data density lower because it results in longer life, fewer errors. # of platters is then simply to keep that lower data density but increase performance again, a trade-off.

6) Others have already mentioned that no, more platters do not double or triple the simultaneous I/O.

7) Random access is not tricky to measure at all, the same test as for single platters can be appropriately used. Perhaps you meant random access difference is harder for you to guess, since the entirety of your post seems to have been guesses.


RE: 2.5" Size
By mindless1 on 2/14/2010 1:58:22 PM , Rating: 2
Oops, in #5 above I meant # of platters is ... increase capacity , not performance.


RE: 2.5" Size
By ChrisHF on 2/9/2010 2:55:37 PM , Rating: 2
quote:
I think that should read "low error rate" rather than high.


Correct. 1 in 10E16 == 1 / 10E16 == 10E-16, which is a very SMALL number.


2 Million MTBF.....Really.....
By Obujuwami on 2/9/2010 1:36:31 PM , Rating: 1
MTBF...a standard used by disk makers is not the most reliable thing. If you take the 2 Million hour MTBF and apply it to one drive, that drive can run for almost 685 years. I find this HIGHLY not possible as the mechanical parts of the hardware will have degraded after a couple of hundred years (lets say 200 just to put a number there). So if this HDD is supposed to run 685 years before failing, but the bearings fail at 200 years, doesn't that make it less than 2M Hours MTBF?

This standard should be scrapped and something else more useful should be put in its place. I think taking the warranty and doubling it would be a good solution, but that's too easy and its not possible to test.




RE: 2 Million MTBF.....Really.....
By jdietz on 2/9/2010 1:41:22 PM , Rating: 2
2 million hours = 288 years. So says google anyway.
2 million hour MTBF = Short term MTBF. Meaning they ran like 2000 drives for three months and got one or two failures. Over time the failure rate goes up. The way these tests are performed aren't public, so scrutinize away!


By Yawgm0th on 2/9/2010 2:17:28 PM , Rating: 4
Really, the test would have to be run until every drive failed to be accurate. That could take a decade or more. However, using hard drives is what causes the majority of failures, not having them powered on. Mechanical parts fail far faster than circuit boards.

A more useful standard would revolve around actual I/O until failure. Something like bytes written before failure (MWBF - Mean Writes Before Failure?) would be useful. Setup a large group of them doing constant writes until a certain percentage of them fail. I'd say 20% is good. This would still take a long time, but not years. I'd wager that 24/7 constant writes should get them to fail relatively quickly.


RE: 2 Million MTBF.....Really.....
By Davelo on 2/9/2010 2:27:20 PM , Rating: 2
Math fail

It's 228.31 years. Still a long time. And to the guy complaining because many drives won't make it that long, MTBF is an average. No guarantees.


By HotFoot on 2/9/2010 2:15:04 PM , Rating: 2
MTBF is not directly related to expected service life, but perhaps the problem is that the service life isn't stated. A statement like MTBF of 2 million hours within a service life of 5 years would be a more informative statement.


RE: 2 Million MTBF.....Really.....
By Slaimus on 2/9/2010 3:25:46 PM , Rating: 2
I agree. With the way MTBF is currently measured, it is only useful as a measure of the DOA or near DOA (fails within a few weeks) rate.

It is an arbitrary pool of disks that are run, and their cumulative running hours before one fails is averaged.

If you have a drive that will never fail within a year, but always fail right after a year, it can have a infinitely high MTBF rating using this method by increasing the drive pool size.

For these numbers to make sense, there needs to be fixed disk pool size, and the pool(s) should be run until 20% (as an example) of the drives have failed, however long that takes. Using that data, you can get a good estimate of how long a single drive will last.


By HotFoot on 2/9/2010 8:03:06 PM , Rating: 2
Downside of that, of course, is it would probably take 4-5 at least years to reach 20% failure for these drives... so... that's not going to be too useful to the consumer either.

I think MTBF is a very important figure for companies with server farms. When you have thousands or tens of thousands of hard drives, this number begins to translate into the bottom line of your maintenance costs.


RE: 2 Million MTBF.....Really.....
By MrFord on 2/11/2010 2:03:31 PM , Rating: 2
Funny thought... if you would try that with a couple hundred Maxtor DiamondMax, the crappy slim 5400 and 7200RPM 20-40Gb IDE drives sold back in 2004, you could probably calculate the MTBF in under a week. They didn't need 200 years to die, that's a given.
http://www.computerpartsdirect.us/images/products/...


By Acanthus on 2/9/2010 3:21:09 PM , Rating: 2
OEM's define MTBF as the mean time before failure if you replace the drive at the end of its recommended service life over and over.

To put this in context. Seagate may still expect you to replace the drive every 3 years. If you do replace the drives on schedule, you will on average have a failure once every 2 million hours of operation.

That is how MTBF is defined by the OEMs. It does not by any stretch of the imagination mean that a single drive will run for 280 years.




By mindless1 on 2/10/2010 2:39:39 PM , Rating: 2
NO!

That is NOT how OEMs define MTBF, and it is not how Seagate does.

Please stop making things up.


Finally!
By twhittet on 2/9/2010 1:29:09 PM , Rating: 2
I've been waiting for such an announcement for 6 months - took them long enough.




consumer
By raphd on 2/9/2010 4:47:49 PM , Rating: 2
they should implement some of that 2M tech in their consumer hard drives. 3 drives failed before they made their birthday.




MTBF
By Oregonian2 on 2/10/2010 8:52:25 AM , Rating: 2
As to what "MTBF" means, there also are some MIL specs and I recall some other standards based specs that define it. Spec usually is referenced. Note too that the little circuit board and it's components figure into the spec as well for the overall drive, in addtion to any internal head driver IC's, etc.

MTBF's can be tricky to use and even more tricky to calculate, but it's a lot better than nothing or just an manufacture's marketing "we're reliable!". :-)




“And I don't know why [Apple is] acting like it’s superior. I don't even get it. What are they trying to say?” -- Bill Gates on the Mac ads











botimage
Copyright 2012 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki