STORAGE

  • 03/19/2014
    3:00 PM
  • Rating: 
    0 votes
    +
    Vote up!
    -
    Vote down!

Last Gasp For Hard Disk Drives

While solid state gains ground, vendors surprise the market with new hard disk drives. Does anyone really need them?

A new, a pint-sized, 15,000 RPM hard disk drive has just been released to the market. It's a fast little product from Western Digital's HGST that is a capacity match for traditional enterprise 3.5-inch drives. Fitted into only a 2.5-inch form factor, however, the Ultrastar C15K600 boasts more than double the performance of existing drives.

The product doesn't come with a Fibre Channel interface, so it isn't a complete throwback to the good old days. But it ships without onboard flash or any of the other new industry tricks, such as helium filling.

The question is, "Who needs it?" That's not pejorative, but this drive goes against the current wisdom of the disk drive business. It's generally accepted that superfast SSD or flash arrays will overtake the market for enterprise performance storage -- a process that's already happening.

In a press release announcing the new Ultrastar, HGST cited guidance from IDC indicating that the demand for 15K drives in 2016 would be twice that for SSDs. in 2016. In addition, HGST said:

First, our customers continue to use 15K HDDs along with a complement of SSDs in tiered pools of storage, depending on their performance, capacity, and power efficiency requirements. Our new Ultrastar C15K600 fills a need for storage that is more cost effective than SSDs, and has the performance, reliability and capacity requirements needed by mission-critical enterprise applications. Also, we see that the industry is transitioning away from 15K 3.5-inch hard drives to smaller 2.5-inch drives to help reduce space requirements, while offering comparable capacities to the legacy 3.5-inch 15K products.

This seems to suggest that legacy arrays are driving the need for the new product, but 2.5-inch drives generally don't fit arrays and need physical adapters. The last sentence suggests new 2.5-inch arrays could be a major factor in the opportunity.

Looking a bit deeper, the legacy story has a few holes. Arrays tend to be homogenous in drive types, and enterprise drives have had long lifecycles as a result. Certainly, the ability to get doubled performance might be an argument for upgrading all the drives in an array, but given the alternatives of SSD and all-flash arrays, it isn't very strong.

The tiering comment is also a bit off-key. Creating a current-view tiering involves a solid-state front-end with bulk spinning drives behind it. Flash/SSD products provide the IOPS, while the HDDs provide the terabytes for archiving, backup, and low-usage data. There doesn't appear to be a use case for SSDs and fast (but expensive) enterprise 15K HDDs, since any such case is better resolved with more SSDs.

Drives like this could give life to new large capacity arrays with, say, 60 drives in a 3U cabinet. A 36-terabyte box like this would deliver just 18,000 IOPS, which is far below the slowest single SSD on the market. We must remember that we have been conditioned by 30 years of performance stagnation in HDDs to think only of capacity when it comes to choosing drives. 

So how much does the capacity really cost? Terabyte consumer grade multi-level cell SSDs retail for around $600. De-provisioning them to 600 GB provides an excellent wear life. While HGST hasn't publicly announced a price for the new Ultrastar drive, based on similar enterprise drives it won't be much cheaper than $600. Of course, the OEM vendors may choose to set premium prices on SSDs for their arrays, and having the new HDD as an option allows for more pricing games. That seems a bit like a bait and switch, with SSD coming out the winner. 

In sum, the use cases for this type of drive aren't obviously strong. It will likely find a market in Japan, for example, where conservatism is king, but SSD is running a better race elsewhere.

The new drive is nice example of design in its class, and a tribute to its design team. Market success would be well deserved, but the deck may be stacked against it. Don't be surprised if we look back in a year or two and say, "This was the best -- and the last -- enterprise HDD."

State Of Enterprise Storage Solid state alone can't solve your volume and performance problem. Think scale-out, virtualization, and cloud. Find out more about the 2014 State of Enterprise Storage Survey results in the new issue of InformationWeek Tech Digest.

 


Comments

SSD for life

Since switching to a laptop with SSD storage, life is much better. I see no reason to go back to a hard drive. SSD startup performance alone is reason enough to abandon hard disks.

Re: SSD for life

Thomas, I did the same. There's no going back!

Re: SSD for life

I agree. Not having a vulnerable HDD in a laptop means one less thing that is liable to fail. You had a long, noble run, but RIP HDD.

Death of HHDs highly exaggerated

In portable devices, of course SSDs will replace spinning hard drives with their glutonous battery consumption. But predictions of the immediate demise of the HHDs elsewhere in the data center is still premature. SSDs can replace HDDs, but that doesn't mean they will. Inertia and unexpected continued use cases have slowed such predictions in the past.

Re: Death of HHDs highly exaggerated

As a fast, enterprise storage device, the HDD is being supplanted by SSD. There's no reason not to. Prices of SSD are crossing over enterprise HDD pricing, while performance is much higher.

Bulk HDD will likely stay around a while, though, with Google's latest pricing as a sign of things to come, they may be all in the cloud in a few years. These are slow, capacity focussed drives, not 10K and 15K spinners.

Re: Death of HHDs highly exaggerated

Jim, your post was a bit sensational IMO, and I'm glad you corrected yourself in the comments to some extent.

Since HGST themselves make both Enterprise HDD and SSD, and declare in their press release that the market prediction for this class of enterprise HDD is 2.5* the enterprise SSD market in 2016...they've made a shrewd decision to release the drive. I would predict they will do well with it

The disk is outstanding with 12Gb SAS, 100 Megabyte ram cache, FIPS 140-2 and ICE/SED drive options. The application is for SAN/NAS where the design is limited to network throughput speeds anyway, and where SSD'd don't necessarily provide a big performance benefit.

The press release also tells you about a size and interface matched SSD to allow mix and match designs and since HGST sell storage, they were certainly right to ensure they had the best mix of capability available in this market (2.5" SAS storage).

Your last comment was right though......HDD's will be with us for some time, especially as we wait for the price of this class of SSD'd to come down. However we need to make sure we don't through the baby out with the bathwater, these drives can in no way be compared to the consumer SSD market.

Re: Death of HHDs highly exaggerated

I think you are jumping from"drives will be around a while" to justify "enterprise drives will be around a while".

SSD are price competitive and performance-wise much faster than even this new enterprise drive. A choice to use enterprise HDD doesn't seem logical!

As for the comment about SANs, which would you prefer? A hard-drive with a 6 millisecond avarage latency or an SSD with just 50 microseconds?

Re: Death of HHDs highly exaggerated

Since the HDD we are discussing is an enterprise drive, and since I was discussing that drive, the comment that HDD will be around for some time to come is implicitly enterprise and not consumer.

C15K600 has about 2mS average latency, and max 7.7mS Seek and R/W with 290MB/s transfer rate to cache, 175/250MB/s R/W sustained Max IOP's about 600.

Retail $700

http://www.hgst.com/tech/techlib.nsf/techdocs/0F76ED89D13F7DA688257C8900...$file/USC15K600_OEMSpec_v1.1.pdf

A comparison SSD might be:

SSD800MH MLC

Command overhead 45uS but Response time is variable from 100uS to 20mS, dependant on Flash firmware overhead. So total times are from 145uS to 20mS

Sustained data transfer 1001Mbps ....so 125MB/s (though for short R/W the IOPs are crazy at 106k/s max for 8kB/32QD on the single port and 21k/s on dual port) High IOP's of course directly dependant on QD being >2

Retail $1442 for 200GB...so I need three of these to get the same capacity as the C15k600

http://www.hgst.com/tech/techlib.nsf/techdocs/428E24D6C2FA344A88257BFE00...$file/USSSD800M_Spec_enc_V1.2.pdf

So to answer you question on which I would choose....it's complicated, and there are places where the advantage of the SSD would be hidden behind network and caching controller times. This drive still has it's place for the next few years. 

Re: Death of HHDs highly exaggerated

As I read the data posted, the C15K600 has an average access time (seek+latency) of around 4 milliseconds, which translates into 250 truly random IOPS, measured over a full drive. Benchmarks on hard drives consistently give better reults, because the test file size is much smaller than the drive capacity.

This is also the center point of the seek distribution statistically. Some IOs will take 12 milliseconds, some will be very short, in the millisecond range.


An SSD average seek typically is around the low end of 20 microsecond, since the distribution is heavily skewed towards the low value. Excursions to the high end are rare.

I would choose a different SSD for comparison. Expensive "Enterprise" SSD are not necessary, especially at the price quoted. A more modern drive is the Samsung MZ-7TE1T0BW, priced on the Internet at $599.99. This unit has 1TB, is MLC like the one described in the previous post, and achieves 98,000 IOPs. Using the industry-standard calculation (sum of all seek times divided by number of seeks) for average access times, this gives an AAT of 10 microseconds.

I guess that's the reality of SSD. I think a Terabyte for $600 with 10 microseconds AAT trumps 0.6 Terabytes for $700 with 4 milliseconds AAT! And the usual FUD about MLC having write wear-out problems doesn't cut it either. If I format the Samsung at 600 Gigabytes, allowing an extra 400 GB of spares, it will easily last 8 years in any use case except an artifical test.


Finally, the NAS use case either uses slow bulk drives, in which case 2TB for $79 on the Internet seems appropriate, or it needs fast drives, which would be SSD. This latter is an ideal SSD use case, since the write percentage of IOPS is typically very small. You could probably format the Samsung to its full 1TB.

Re: Death of HHDs highly exaggerated

There seems little point in continuing since you are now replacing (comparing) enterprise SAS drives with consumer SATA drives which are essentially impossible to configure in large PB arrays.

Just in passing, the Samsung drive (a great laptop drive I agree) is heavily asymetric with the write throughput and write IOPs only achievable with the write behind RAM cache enabled (main memory in the PC), otherwise the throughput is essentailly half or less the read throughput.

...and while overprovisioning certainly can extend the drive lifetime write amplication and wearout are real problems for MLC. The SSD800 had a 36PB write lifespan....and I'd assume the Samsung would be close to one tength that before oversprovisioning.

Re: Death of HHDs highly exaggerated

With "consumer" SSD becoming so inexpensive, and large, it's hard to argue for the benefit of "Enterprise" SSD. Sure, they are a bit more durable, but overprovisioning is the reason for good wear life, and my suggestion for derating a 1TB drive to 600GB will address that.

As for dual interfaces and all that stuff, the benefit when both interfaces are in the same control chip on the SSD is questionable. Arguments that dual-porting allows redundant paths are rather weak, given that anyone that worried about data availability needs appliance level redundancy not drive level. This is especially true in the NAS space.

 

Re: Death of HHDs highly exaggerated

Seriously?? I had no idea you could use SATA in the same sentance as "enterpise class reliable" storage. Things must have changed a lot in the couple of years I've been retired.

IMO SAS allows:

Hundreds of drives per controller which is important in PB stores.

Data path redundancy which is critical to multi drive reliability

Multi Host connectivity

Now mixing SAS and SATA drives (both HDD and SSD) within a design may be advantageous depending on the relaibility and throughput required of particular data, but multi port SATA controllers are expensive and low throughput compared to SAS controllers in my experience.

...and to me...."all that stuff" is seriously important.

We'll just have to agree to disagree.....

Re: Death of HHDs highly exaggerated

I think you were buying the FUD on SATA! I've built large SATA array and they worked just fine. One of them was 70 Petabytes!


Let's call it a day and let someone else comment.

Re: Death of HHDs highly exaggerated

Like all evolutions of storage media, HHD will take its place in the capacity/speed/cost equation.  Folks with budgets, without the need for super speed will go the cheaper route.

Re: Death of HHDs highly exaggerated

HHD's are a very interesting technology. Having worked on the design of HHD drives and the OS software support for them they can have some very interesting impacts on large data store design. 

IMO though, they will die out as a single unit entity and be replaced by (at least initially) SAS SLC SSD's paired with lower speed high capacity HDDs in arrays. For most classical mixed 70/30 workloads hybrid write structures can provide all the benefits of all SSD equivalents that cost much more. Though the dual physical unit is a risk, it's fairly low risk that separation would occur in a properly controlled environment. Dual unit also eliminates write wearout with a sync and replace SSD strategy.

Re : Last Gasp For Hard Disk Drives

@ Charlie Babcock, I can't agree more with you on this. As you have mentioned data centers where it is difficult to predict the demise of hard drives, I would like to add another thing that is developing countries. Technology is seen to trickle down to developing countries quite slowly. This factor might keep the hard drives in business for quite a bit of time in future.

Re: Last Gasp For Hard Disk Drives

There a couple of other things we have to keep in mind while predicting the demise of HDDs. There have been predictions about the demise of desktop computers as well for quite some time now. But desktop computers have proved to be resilient and most of them use HDDs that will keep HDDs in business. Other thing is the cost. SSDs are still costlier than HDDs and not everyone can afford them.

Re: Last Gasp For Hard Disk Drives

@SachinEE, which is costlier depends on your yardstick. SSD might cost more per Terabyte, though I'd argue that we are nearing parity between entrprise HDD and MLC SATA SSD, but on a $/IOPS basis, one SSD is worth a couple of large arrays of HDD, which is a phenomenal price-performance edge.

Capacity isn't everything, and tiering and caching mean that slow bulk HDD and an SSD top tier will deliver much better performance than a farm of HDDs will.

Re: Last Gasp For Hard Disk Drives

As you said: "Capacity isn't everything, and tiering and caching mean that slow bulk HDD and an SSD top tier will deliver much better performance than a farm of HDDs will."

Tiering, caching and the need for immense data stores will give HDD life for many years to come in those areas that don't need huge IOPs.

 

Re: Last Gasp For Hard Disk Drives

There is a point where the low IOPS of HDDs becomes a major issue even in bulk applications. We've seen a decline of IOPS/GB from around 15 (9GB 15K RPM) in the 1990's to today's 0.03 (5TB SATA drive).

The implication is data is much slower to get to. With MLC solid-state drive prices declining 20 percent or so annually, I predict IOPS/GB will cause a general transition from bulk HDD to bulk SSD in a few years time. Issues such as low power usage and high packing density will add to the attractiveness of a technology that can get 15 or more IOPS/GB.

One reason Google is a major SSD manufacturer (consuming internally) is that they are already well into that transition.

Re: Last Gasp For Hard Disk Drives

Well said: " I predict IOPS/GB will cause a general transition from bulk HDD to bulk SSD in a few years time. Issues such as low power usage and high packing density will add to the attractiveness of a technology that can get 15 or more IOPS/GB."

Therefore there is still a window of opportunity for (particulalry enterprise) HDD's for the next few years....which get's back to the original point of the thread .

I don't think anyone disagrees that SSD's (based on whatever final technology is used) will eventually be all there is and HDD's will be a thing of the past. We're not there yet though, and there is still some market left to leverage.

Re: Last Gasp For Hard Disk Drives

@JCreasey, it seems what I've said is being misinterpreted. To restate:

1. SSD are much faster than enterprise HDD. This means the window for these HDD is closing fast, and SSD will replace them completely.

2. Bulk HDD are very slow. With MLC SSD prices falling rapidly, these solid state drives will begin to replace bulk hard drives in a few years.