Yesterday Nimbus Data Systems announced the release of their S-Class storage array. What’s different about this storage device is that it uses purely NAND memory rather than traditional spinning disk. With it’s arrival, we’re seeing another nail in the coffin of the traditional hard drive.
OK, so the S-Class isn’t going to replace hard drives overnight – at $25,000 for 2.5TB of storage (or a hefty $10/GB) it seems expensive. But in reality it isn’t. The $10/GB mark is about what you’d pay for tier 1 Enterprise storage and for that you get a device which is lower on power, footprint and offers performance far exceeding traditional tier 1 arrays.
Nimbus are clearly looking to fill the gap between Enterprise arrays that have shoe-horned in SSDs (EMC and others) and the super-fast super-expensive arrays produced by the likes of Texas Memory Systems (RamSAN). This continues to blur the boundaries between disk and memory and in the short term meets customer requirements for faster storage, overcoming the limitations the hard drive.
However, the future isn’t memory-emulated hard drives. The future may not be drives at all, at least for main memory. After all, what’s the point forcing the storage and retrieval of data in an artificial construct like a hard drive with SCSI? HP recently disclosed the results of research into memristors, making it easier and cheaper to make memory “with memory”, that is a cheaper and more dense version of today’s Flash technology. As virtualisation moves on, SSD & memristor based technology will form the basis of storage used for primary data in virtualised environments. In fact, I’d postulate that hard drives will get slower over time, as they are relegated to more of a secondary storage medium, holding inactive and replicated/backup data.
The use of memory as permanent storage also removes the need to store and retrieve data in the hard drive format. Today’s SSDs simply emulate hard drives, but it isn’t essential; look at the Fusion-IO products. These sit on the server bus and offer orders of magnitude faster response times than hard drives can – or even storage arrays with SSDs.
The hard drive isn’t going anywhere overnight (witness the crash of STEC shares and EMC’s EFD stockpile), however storage technology is changing as we move into a new era with virtualisation and cloud computing. Hold onto those old hard drives, they will be a museum piece one day.
14 Responses to Enterprise Computing: The Slow Demise of The Hard Drive
Leave a Reply
You must be logged in to post a comment.
- Use Symantec and know your sensitive data is protected with industry-leading backup & recovery software.
Experience Symantec Backup Software
[...] This post was mentioned on Twitter by Michel van Kessel, LG Stream. LG Stream said: Enterprise Computing: The Slow Demise of The Hard Drive http://j.mp/dmadud [...]
Social comments and analytics for this post…
This post was mentioned on Twitter by lgstream: Enterprise Computing: The Slow Demise of The Hard Drive http://j.mp/dmadud...
On that subject, also think NetApp’s Flash Cache; it’s not SSD, and sits on the array bus. The bandwidth is orders of magnitude greater. Emulating disk to act as a cache; now, that’s madness!
The missing piece of this discussion is Semiconductor economics. The cost of Fabs is immense and you need to build a LOT of fabs to build enough GIgabytes to replace a large percentage of enterprise hard drives. It will make sense to replace FC or SAS, probably makes sense to do that now! But bulk storage on SATA will be with us for a long time to come.
Chris,
There is hardly a day that goes by that a post like yours above is available to even casual tech readers. I respectfully submit that your post is not much different than all the others in that it omits a real effort to put a time line on the transition you forecast. You also fail to mention hard drives in the context of applications in client computing (laptops, desktops) or in consumer electronics,or in back-up. All HDDs, I assume, are referred to in your conclusion as soon-to-be museum pieces. I am forced to assume that you mean within your or the reader’s lifetime to which my reply would be “so what”. In addition, your reference to memsistor research refers to an entirely different technology than the technology behind SSDs today, a point that I feel is germaine within the context of your post. Finally, to suggest that Nimbus’ has announced a product that is on parity with the cost of HDD based enterprise systems is simply not true. How about a fair apples-to-apples comparison? Let me fill in some blanks for you. Enterprise class HDDs (meaning SAS/SCSI) sell for about $1.30/GB average sales price (ASP) net to the HDD mfr (based on CQ1 data). The ASP for HDDs like Western Digital’s VelociRaptor with 6 G/bs SATA, rotating at 10K RPMs, with a 3ms latency, is far less. Current forecasts for NAND and magenetic HDD density and price suggests that the cost variance between them will be maintained at least within the forecast period commonly used by IDC, Gartner, iSupply, and Forrester (5 years). With updates due from IDC and Gartner soon on the comparisons between HDD & SSD I suggest you reference them. IDC made a presentation in March this year that reflected a 2010 forecast of approximately 500K enterprise class SSD shipments versus 40 million enterprise class HDDs.
I think the product announcement from Nimbus is very important. I think that NAND based SSDs have a bright future in the storage hierarchy. I think that much of what has been written about SSDs relegating hard drives superfluous is simply not useful.
[...] This post was mentioned on Twitter by sfoskett, Chris M Evans, Juan Carlos Sánchez, Tyler Romeo, cfcassidy and others. cfcassidy said: RT @SFoskett [Shared] Enterprise Computing: The Slow Demise of The Hard Drive http://bit.ly/c9IDdb [...]
Alex, of course the Netapp PAM solution isn’t perssitent storage, merely an acceleration for reads, so you can’t consider it similar to the Nimbus hardware….
Chris
Dave,
That’s a fair point. I’ve no idea what the capacity is for increasing production of NAND memory; presumably the ‘big boys’ like Seagate are well into this already.
Cheers
Chris
Gary
Agreed, I haven’t put a timeline into the forecast. Let me go away and think about that in more detail. Regardng cost, I’m referring to the price of systems $/GB rather than individual hard drives. The $10/GB for the Nimbus system is comparable to a tier 1 enterprise array price.
As far as laptops and other devices go, we’ve already seen SSD technology in laptops, iPads etc. This is well established and the replacement of HDDs in those devices is simply a matter of time – when the cost makes sense. Let’s face it, today the consumer with money can already choose to use an SSD in his/her laptop. I’ll make a prediction; the standard permanent storage device for laptops in 5 years will be SSD. We won’t be using HDD by then. Feel free to challenge me on that in 5 years’ time.
My “so what” comment was to highlight how our view of storage today is that data is packaged and accessed in terms of a hard drive. This, in SSD terms is really an artifical construct meant to simulate existing technology. I think we’ll quickly see a change away from SSDs emulating HDD and a more direct integration of persistent storage with the server.
BTW, I did class this post as “enterprise computing” rather than discussing the consumer side of things, however I appreciate your comments and views, they are always welcome.
Regards
Chris
[...] This post was mentioned on Twitter by sfoskett, Chris M Evans, Juan Carlos Sánchez, Lasse Christiansen, Tyler Romeo and others. Tyler Romeo said: Enterprise Computing: The Slow Demise of The Hard Drive http://ping.fm/5Qrt4 [...]
Agreed, this is a good subject to revisit in 5 years. However, I wanted to hasten to say that my “so what” comment was meant to imply that I agree, and in fact think it is a ‘no-brainer’, that HDD will eventually be replaced one day. However, I do not think the time frame is knowable at this point. The amount of infrastructural change we need to make just to break free of current bus structures, for example, requires new everything. How we get to a point where hardware and software designers agree on that is a major uphill battle as I contemplate it. We curse the chains of the IBM/X86/MS-OS chains already but I suspect that the weigtht of those chains will hang heavier and heavier, much like on Marley’s ghost. On the other hand a computing world, without any common platform, is worse than any nightmarish scenario I can imagine. However, if forced to, think “Mad Max In The Server Room”. Your comment was “I think we’ll quickly see a change away from SSDs emulating HDD and a more direct integration of persistent storage with the server”. I would simply take away the adjective “quickly”.
By the way, I thik this topic is extensible all the way to Kurzweilian “Singularity” theory. Now there’s a thought….or two…or three…
Best,
G
[...] The Storage Architect » Blog Archive » Enterprise Computing: The … [...]
Colleagues,
I think one very important topic regularly is left out of discussion whenever anyone is discussing SSD vs HDD:
I think it is very safe to assume that for 99% of all requirements we see today, a disk-only or tiered approach will easily suffice.
When discussing the speed of a technology alone, we need to keep in mind that the average IOPS/GB required for the data we keep in general, can be provided by virtually any technology. (stone slates excepted, perhaps…)
Also, in a power consumption per GB, the German magazine iX has recently (in their current edition) calculated that SSD is using MORE power per GB than S-ATA Disks.
Thus I think one should very closely determine the actual needs – and model the storage environment based on that. What is needed for what EMC calls FAST or IBM calls Easy Tiering is a prediction algorithm moving the blocks which are under heavy load over the tiers.
This algorithm needs to be flexibly adaptable to the installation’s specific workload pattern – meaning it needs to be able to learn – and quickly.
To me it seems we’re on the peak (or approaching it) of the hype curve with SSD now – we’ll be seeing a time of diminished interest in SSD before (with the maturity of automated tiering on block level) the technology will become generally accepted and a standard building block of IT infrastructure.
Of course I may be wrong – but I doubt it.
[...] The Storage Architect » Blog Archive » Enterprise Computing: The Slow Demise of The Hard… [...]