Lay-offs took place

I was told by one person anonymously that 900 people were laid off by NetApp in US yesterday, mostly older people. It's not verified.

Chris.

NetApp layoffs insane

NetApp is no longer the best place to work. Their latest layoffs include a number of long term senior CLOUD strategists and Enterprise Architects who were hired to help provide the company with direction and focus on Cloud.

One can only assume, that the entrenched incumbent sales leaders have stuck their heads in the sand, preferring to continue to sell in outdated modes, via storage based solution sales, rather than helping their partners and their channel design and build truly competitive Cloud services. Their ignorance of Service Strategy, Design, and the basic tenets of ITIL and ISTM will well serve their competition well, and white box storage behind well designed Cloud Services based on Open Source will continue to eat their lunch.

-------------------------

Sent anonymously to me.

HDD vendor way back

Your piece on the death of traditional HDD vendors is interesting, but I wonder if there is a way back for the likes of Seagate and WD? It seems to me that today we have something of a cartel on the supply of flash memory chips, as a result of which we see artificially inflated prices for emerging SSD drives (£350-£450 for a Samsung 840 Pro 512Gb, for example, which is many multiples the price of a top-of-the-range 2 or 3Tb HDD).

This is because the emergent dominant players in this market are currently milking it for all they are worth. It's often the way, and if a market stagnates with 2 or 3 major players, we see little or no commercial pressure to innovate or reduce prices (think Microsoft in desktop, or even say Canon/Nikon in cameras, Nintendo/Sony/Microsoft in games consoles, etc,etc ).

So a possible route back for Seagate and/or WD would be to purchase a Flash Memory fabrication plant and then run out a line of decent quality drives that seriously undercut the existing cartel. Give the market say 1Tb Flash drives for an initial £500 [aiming to drop to £400 when the inevitable fightback begins] and there's the chance of winning back market share.

If they wanted to do this, Seagate/WD could fight back with say last year's SSD technology (one die size large fabbing, slightly lower clock rates) which may enable them to buy up fabrication gear on the cheap. They won't be able to compete with Samsung/Kingston/etc in performance terms, but as long as they offered a drive that was markedly faster than the best HDD and seriously undercut the cartel, they may succeed.

The one thing SSD doesn't have today is a high-capacity drive. Samsung/WS could do it. Question is, after the relatively recent floods in Thailand, and the time, trouble and effort they undoubtedly put into recovering their HDD fabrication capacity, have they got enough left in the tank to scale up a serious SSD challenge. I suspect we'd learn that the flooding is a much larger factor here than merely being caught asleep at the switch...

[Sent to me]

I reckon if CommVault's share price is realistic then EMC is dreadfully under-valued by the market. On the other hand the opposite could be true! I just can't make sense of it.

Chris.

Are Disk Drive Vendors screwed (by flash)?

Read Mike Shapiro's view that HDD vendors are screwed by flash (included bellow). What do you think?

- Is Seagate clawing its way back?

- Is Toshiba sitting pretty with NAND foundry and HDD manufacturing operations?

- Can WD claw its way back?

- Are hybrid SSHDs enough for the HDD vendors enabling them to effectively ignore SSDs?

Chris

------------------------Mike Shapiro interview-------------------

The disk drive vendors have been utterly screwed by mismanaging the disruptive force of solid state drives: that's the view of Mike Shapiro - lately a storage bigshot at Sun and Oracle.

Mike Shapiro was a Sun Microsystems Distinguished Engineer, CTO, and VP of Storage for Sun and then Oracle. He is most recently a founder at a stealthy storage startup about which we know nothing. We spoke to Mike about his views on flash drives and the HDD suppliers.

El Reg:Did the disk drive companies enter the solid state market in a timely manner?

Mike Shapiro: Despite the emergence of STEC as the first enterprise SSD vendor in 2006 (when we first talked to them at Sun) it is remarkable to me that Seagate, HGST, and WD all failed to enter the SSD marketplace until 2011 (5 years later, by which time it was essentially too late). And when they did, they did the obvious dumb thing of pricing the SSDs above the 10k and 15k RPM drive lines - i.e. they made the classic error of thinking that they could solve a disruption by just organising their own revenue streams regardless of other market forces.

El Reg: So what happened?

Picture of Mike Shapiro, Bryan Cantrill

and Adam Leventhal

Mike Shapiro:As a result of this massive screwup, a raft of solid state drive companies entered the market, which in turn I believe spurred the NAND vendors to say, 'gee, if that's all it takes to mark up our NAND, surely we can do a better job of that'. Furthermore, the NAND companies are entirely staffed by ex-DRAM people whose major lesson in life was that DRAM was killed from a margin point of view by letting someone else (the x86 CPU) commoditise the interface to it. So the idea of keeping the deep NAND interfaces secret, building their own controller or firmware or drive, while doling these secrets out very cautiously, made sense.

El Reg: Can you split SSD market development into phases?

Mike Shapiro: So we really see three phases of the SSD market using the rear-view mirror -

(1) STEC creates the market (first customers EMC and Sun)

(2) Startups enter the market, partnered with the NAND suppliers

(3) NAND suppliers become SSD suppliers, kill off startups (and STEC)

El Reg: How did the HDD suppliers mis-read things?

Mike Shapiro: How different it might have been if they (HDD suppliers) had acted in stage (2) Furthermore, the disk vendors assumed that all of their volume (i.e. small servers and laptops and desktops) would come from the 2.5in disk drive form factor for client [products] and would become the dominant form factor around 2009-10. But instead, thanks to cloud and tablets and iPhones, that entire transition has in fact been killed - the server market is in decline. All of the mobile computing and devices use 100 per cent flash, and so in fact the remaining use case for disks will be 3.5in (bulk storage).

El Reg: Bulk storage disk drive sales prospects look okay though?

Mike Shapiro: [Yes] thanks to the hyperscale customers like Google and Facebook and Apple, the market for bulk disk is now a direct sale (i.e. literally directly to the end customer) rather than an indirect one (through HP, Dell, IBM, Oracle etc). So we see an ability for the HDD guys to (temporarily) grow margin as they adapt to this opportunity, yet over the long term the opportunity to keep their position in the volume client storage device business has been entirely squandered.

El Reg supposes that Shapiro's criticisms are directed mostly at Seagate and Western Digital. The third HDD supplier is Toshiba and it operates flash foundries in partnership with SanDisk. Seagate has just widened its flash storage offering with three SSDS and a PCIe card powered by Virident, and Seagate is now almost 10 percent owned by flash foundry-operating Samsung. WD has an investment in all-flash array startup Skyera and is expected to widen its SSD range soon.

Can Seagate and WD catch up? Shapiro would think not. They are utterly screwed. ®

Why have CommVault shares outperformed EMC's?

Although CommVault and EMC have both exhibited steadily climbing annual revenues and profits CommVault shares have outperformed EMC's. Why? Are investors and the stock market commentators, analysts and influencers crazy?

A coming story about CommVault's results has charts that show the two company's annual results and their share price movements compared. How is it that their share prices have moved so differently?

From Paper's author Mario Blaum

Chris,

in your note you write: "The paper, by IBM Almaden researcher Mario Blaum, professes to solve a problem where RAID 5 is insufficient to recover data when two disks fail."

This is an incorrect statement. I never say that in the article. RAID 5 cannot recover data when two disks fail. For that you need RAID 6. The problem addressed in the paper is one disk failure and up to two "silent" failures in sectors. That is, when a disk fails and you start reconstructing using RAID 5, you may find out that you cannot read a sector, and then you have data loss.

In order to handle this situation, people use RAID 6. But this is a very expensive solution, since you need to dedicate a whole second disk to parity. For that reason, we developed the concept of Partial MDS (PMDS) and Sector-Disk (SD) codes, which have redundancy slightly higher than RAID 5 but can handle the situation described.

Let me point out that Microsoft researchers in the Azure system came up independently with a similar solution, though the application was different (our main motivation was arrays of flash memory). Both the IBM and the Microsoft solutions involve computer searches. A theoretical solution was an open problem, which I provide in the paper you mention. The solution involves mathematical concepts like finite fields (to which you refer ironically as a mathematical side line with no real world applicability). I will make no apologies for the math and you are certainly free to believe that this is just a mathematical curiosity. However, we recently presented our results at FAST13 (Plank, Blaum and Hafner) and there was great interest. Jim Plank and I are preparing an expanded version under request for ACM. Best regards.

-Mario Blaum

Posted for Grant from US Army

Very simple answer concerning the RAID 5 paper -

The math is just to validate the findings and is immaterial to his basic premise, which is that by adding additional parity bits to a RAID 5 array, the fault tolerance of RAID 5 exceeds RAID 6, and at a much reduced cost.

-------------------------

Chris comment: Wow; I wish the abstract had said that!

Comment from Forum member who has forgotten his password

Basically the paper gives the theoretical proof and underpinnings for SD codes. SD codes are more efficient than RAID-6 in that you do not have to dedicate 2 disks for parity which tolerate the failure of one disk and one sector. ie RAID-6 protects you from a disk failure and an URE (unrecoverable read error) during a RAID rebuild.

With SD codes, a disk and a sector within a RAID stripe are dedicated for parity which is more efficient and actually maps to how devices fail. ie Entire disks rarely fail, what is more likely is the failure of a sector within the device. The USENIX paper quoted in the comments has more information. A key quote is below

"We name the codes “SD” for “Sector-Disk” erasure codes. They have a general design, where a system composed of n disks dedicates m disks and s sectors per stripe to coding. The remaining sectors are dedicated to data.

The codes are designed so that the simultaneous failures of any m disks and any s sectors per stripe may be tolerated without data loss"

The research paper offers the proof for why this is so.....

-------------

Chris comment - perhaps we need erasure codes to recover forgotten passwords....

Crowd-sourcing interpretation of IBM RAID 5 extension paper

The thread for comments describing and interpreting and reviewing IBM Almaden Researcher Mario Blaum's paper: "Construction of PMDS and SD Codes extending RAID 5" which can be downloaded as a pdf from here.

My insufficiency of cerebral matter prevents me so doing. Help please.

Chris.

"As a matter of interest, where did you get the idea that "Every byte stored in Amazon's cloud, or Azure, or the Googleplex, or Rackspace, is a byte not stored in a private VMAX, VNX, Isilon, FAS-whatever, VSP or HUS, StoreServ, StoreVirtual, StoreWhatever, V7000, XIV, or DS-whatever."?

Because it seemed self-evident. Am I talking rubbish?

Chris.

Cloud storage & legacy storage supplier vertical disintegration

This topic is for comments to the notion that public cloud storage growth will cause legacy storage product sales to collapse with existing storage suppliers becoming cloud storage service operators, if they can, or cloud storage service component suppliers, They will have to vertically disintegrate.

LTFS and ugly ducklings

If it quacks like a duck, looks like a duck and swims like a duck then it's a duck, and not a swan. So what is LTFS?

LTFS is a way of providing file:folder type access to files on tape using drag and drop operations. You are no longer forced to use a backup application or equivalent software to move files to and from tape and so, the story goes, an obstacle to wider tape usage is removed.

My query is; how much of an obstacle is it? If I, as a user, have file:folder access both to an external disk and to a tape drive; on which device will I store my files? It will be disk, natch, because access is faster and there will be a backup, probably also on disk, in case disk numero uno goes tits up. Tape is still tape, still slow compared to disk, even inside an LTFS wrapper.

If the company I work for already has a tape system and it implements LTFS then yes, I could use the tape but why would I want to do that? If I was forced too then, fine, reluctantly I'd use the damn thing, but sneak in USB sticks to make life easier where I could.

There's a rumour that one large tape system-supplying vendor has not one LTFS-using customer in Europe. It wouldn't be surprising. For everyday access to files, putting LTFS access on tape in a disk-using world, is like putting lipstick on a pig. It's still a pig.

Am I right or am I being a dickhead about this and missing a point or points?

What is the real problem here?

Sent anonymously to me:-

A Reg reader has the following comments to make on the story Rise Of The Machines: What will become of box-watchers, delivery drivers?. The request to send this message came from the IP address 94.211.113.82.

This is more or less a transitory problem. Even if this now causes 10M people extra on welfare, the problem will go away in wotsit 55-odd years. Of course, maybe there'll be no state in 55 years.

And that's assuming there really are no alternatives. With the screaming about needing more foreign workers, well, maybe this pool of labour can fill that need, who knows.

That there's little manufacturing on US soil left, I can't really be arsed to care. Partially their own fault for letting that happen. Then again, as cheap becomes popular it becomes more expensive, making manufacturing elsewhere interesting again. There's plenty of room for innovation here.

I don't think we should frame it as an insurmountable problem. There'll be change, and it'll be painful, to be sure. But with a little looking forward for opportunities instead of problems, a lot can be done.

Also, google has already lost all credibility as "doing no evil". But I still don't think I'm going to be enamoured of the idea of holding back true technical invention for the sake of the incompetence of the representatives failing to care for their large pool of workers.

Besides, it's a wider problem, much wider. Both in geographical sense -- foxconn building factories run entirely with robots, nary a human in sight -- and the technology sense. You already pointed out the Luddides (and they did have a point), and we've been doing nothing but putting people out of work.

Alright, not entirely true. We've also created a lot of cubicle potato jobs, both as "knowledge worker" (arguably positive) and as what looks like machine minders but really are minded-by-machine drones, barely thought capable of clicking an icon.

And that, that is much worse, for it makes us string puppets of our own technology. The same is happening in a lot of security applications, but the redmondian desktop is perhaps the most widespread insultingly patronizing string puppeteering of human workers available today.

But anyway. We've automated the shit out of many a thing, putting ever more people out of work, and in the meantime the population has done nothing but grow.

What, now, is the real problem here?

Facebook's OCP is unrealistic - for the rest of us

Can Facebook's OCP vision succeed in turning back time and disaggregating the IT server, storage and networking industry into separate component developments linked by support of common interfaces?

The future of storage

I've been talking to Jean-Luc Chatelaine, EVP strategy & technology for DataDirect Networks, and I'd like to check out his view of things with you; it being surprising.

He thinks that, starting 2014 and gathering pace in 2016, we're going to see two tiers of storage in big data/HPC-class systems. There will be storage-class memory built from NVRAM, post-NAND stuff, in large amounts per server, to hold the primary, in-use data, complemented by massive disk data tubs, ones with an up to 8.5-inch form factor and spinning relatively slowly, at 4,200rpm. They will render tape operationally irrelevant, he says, because they could hold up to 64TB of data with a 10 msec access latency and 100MB/sec bandwidth.

He claims contacts of his in the HDD industry are thinking of such things and that it would be a disk industry attack on tape.

-------------------------

So .... what do you think of JLC's ideas?

Wrong interpolation

Sent to me anonymously and posted here as the quickest way to get the comment known:-

In your story, "Are you ready for the 40-zettabyte year?", you write, "The amount for 2020 would be 43.2ZB by interpolation." The increase is doubling every 2 years so you cannot use linear interpolation.

Half way between 2019 and 2021 the data will be closer to the 2019 amount than the 2021 amount. So 40 zettabytes looks reasonable for a continued doubling every two years.

-------------------------

Is he right? Am I innumerate?

Chris

Fixing Dell Storage

Dell storage revenues have declined for two years and its head, Darren Thomas, has just resigned. How should Dell storage be fixed so it delivers on its promise?

Use Steelheads

Did you know that customers worried about latency between their NetApp DC and AWS can deploy Riverbed Steelhead WAN optimisation? All they need is an appliance or software version of Steelhead in the DC and a Cloud Steelhead (CSH) at the AWS side and the NetApp replication is deduped along with any other data directly between DC and AWS. No need at all for the Colo site that NetApp specify and the CSH is set up as easily as buying an instance from Amazon in the first place via an online portal.

Regards,

Jim Morris Group Business Development Manager

Zycko Benelux BV Smart Business centre, Daalwijkdreef 47, 1103 AD, Amsterdam The Netherlands

Sent to me and posted anonymously

What about what IBM does with their object based file system on IBM i? It seems to work pretty effectively at putting the data that will benefit the system most on flash and leaving the rest on the spinners.

Do their patents ( or the radically different nature of the storage management ) mean that no one else can do this? I seem to recall the BeOS file system having some of the same features, too.

--------------------

Chris

Impractical SPEC sfs2008 NFS benchmark win by Huawei

Huawei soared tio the top of the SPEC sfs2008 NFS benchmark ranks with a 3 million+ IOPS score. Previous record-holder Avere didn't think much of it and here's what a spokesperson said:-

Unlike Avere's SPEC posting that used a single namespace (as well as the top results from both NetApp and EMC Isilon), Huawei used 24 filesystems and thus requires 24 mount points for each client. This is completely impractical from a management standpoint since the client has to somehow know which file system to look into for their data. It is like having 24 different internets and having to know that espn.com is on internet #17.

And as the data grows, you need to add new file systems (e.g. 25, 26, etc.) and move data between file systems as they fill up. This causes downtime as data is moved between the file systems and all the clients and application servers need to be constantly updated with the new location of the data. There is no abstraction layer separating the physical from the logical. So, while the solution they provided has high performance, it is just not practical in a real-world scenario.

There are many other gotchas in their test build that make it more of a science project than a real-world system, such as the requirement for two networks, both 10 GbE and Fibre Channel, limited scaling with 16 NAS engines max and more.

-----------------

Sour grapes?

Real point?

Seems a realistic point to me.

Chris.

Pedantry in a humorous vein

Sent to me and I couldn't resist posting it here:-

Dear Mr. Mellor,

I realize it is a bit of a quibble, but shouldn't that be 'about 125 miles', 'a bit under 125 miles', or (most accurately), 'a bit over 124 miles'?

I will restrain my pedantic metric tendencies, and not grumble about the different sizes of mile...

Anon - name withheld by me.

PMT is now an acronym for Pendantic Metric Tendencies :-)

Chris.

Options for Quantum - exiting tape?

With activist investor Starboard Value popping up out of the blue and claiming it owns 15.6 per cent of Quantum's shares and therefore wants a seat on the board and advice proffered about how to raise Quantum's share price taken seriously, what should Quantum do, and what would Starboard Value's advice be?

One suggestion is that Quantum could sell off its Scalar tape library business and exit the tape business altogether. This could enable it to pay off a lot of debt and stop making, presumed, losses from its tape business.

Is tis a good thing to do?

Who might buy the tape business? HP, IBM, Oracle, Qualstar, SpectraLogic or Tandberg Data?

What would a Quantum tape exit do to the Linear Tape Open consortium?

Have at it :-)

HP Memristor prospects

I had an e-mail from Blaise Mouttet re the notion that the memristor could be an important part of any HP recovery. Here it is:-

----------------------------

Chris,

In your story last month "Flashboys: HEELLLP, we're trapped in a process size shrink crunch" [http://www.theregister.co.uk/2012/10/12/nand_shrink_trap/] it was commented by Bennet that "If it works who cares?" in spite of the fact that the memristor models appear wrong. Imagine if Violin Memory tried to manufacture flash memory arrays using incorrect models of transistors. Somehow I don't think that would work out too well for product development. In engineering good models are required to manufacture reliable products. The fact that Bennet does not understand this point is illustrative of either his incompetence or his inability to grasp what the "memristor" argument is really about.

Regarding the comment by Gartner's [Valdis] Filks that the memristor could represent the saving of HP this is probably based on a misunderstanding of HP's patent position. Samsung owns the patent for the TiO2 device HP originally claimed to be a memristor (see US Patent 7417271 - http://www.google.com/patents/US7417271). HP does not have a basic patent for metal oxide ReRAM and most of the metal oxide ReRAM patents are held by other companies (Unity Semiconductor, Panasonic, Sharp, Samsung). Meanwhile Stan Williams basic memristor patent (application 11/542,986 filed in 2006 -http://www.google.com/patents/US20080090337?dq=electrically+actuated+switch+williams&hl=en&sa=X&ei=FOqTULHMDNDU0gG804CwDw&ved=0CDgQ6AEwAg) has been repeatedly rejected by the US patent office and Hynix's relevant patents are almost completely devoted to phase change memory based on chalcogenide materials rather than the metal oxide ReRAM which HP claims to be a memristor. How exactly does the memristor represent the "saving of HP" if they don't have any fundamental patents and their alleged manufacturing partner is devoted to a different technology?

-----------------------------

Is HP delusional over its memristor technology and IP?

Chris.

Oracle RAC on NetApp FlexPods

Here's a mail I received about this story:-"NetApp and Cisco waggle shrunken ExpressPod at Hitachi and friends"

----------------------------

[Re] your article about FlexPod. Although I find Oracle's support of RAC on VM to be a massive step forward I also see it as slightly underhand. They still do not support processor pinning in any other virtual environment other than their own so in my view the support of RAC is negated by still requiring to licence every core in your VM farm or buy a site licence. Where I work we have been forced down the OVM\OVS route for this very reason. We do however run a mixed environemnt and with the current state of OVM\OVS I would much rather stick with physical tin for Oracle and a VMware farm for the rest. Unfortunately due to our requirments to stand systems up rapidly, etc. we have had to go down the OVM\OVS route.

What are your thoughts?

--------------------------------------

Indeed, what are your thoughts? Physical or VMware virtual tin for Oracle?

Chris.

Flash array performance

A supplier ding-dong (disagreement) about flash array performance is going on.

Following a story about Kaminario's K2 flash array reaching 2 million IOPS of performance:-

http://www.theregister.co.uk/2012/10/01/kaminario_k2_2m_iops/

Violin CTO and co-founder Jon Bennet posted a comment rebutting Kaminario's claims and starting out with this sentence: "Kaminario's claims described in this article are so ludicrous and in some cases blatantly false it is hard to know where to begin, lets start with the simplest."

Shachar Fienblit, Kaminario VP of Engineering replied to that, saying "The comments written by Violin’s CTO mix apples and oranges from multiple benchmark data points resulting in false conclusions. Let me clearly and factually present the truth...."

This is all worth a read to understand better the relationship between vendors' claims, benchmarks and customers own experience.

Tracking users across the file system landscape

Is it naive to think that users' file system accesses should not be tracked by business' IT systems in order to safeguard business' confidential information? I find the idea abhorrent, like the existence and activities of the US TSA, but have come round to thinking that's a naive view.

As business should track its staff's file system accesses because their are rotten apples, either through mistakes or malice, in most organisations and they can damage everybody else's interests. Trust, but verify.

Obviously - very obviously - a single pane of glass through which to see all your data protection activities and statuses is a darn good idea. It's surely a no-brainer. Bocada and Aptare are doing everyone a service here.

Chris.

Cleversafe

Cleversafe (http://www.theregister.co.uk/2012/09/03/cleversafe_1tb_sec/) has a 10EB object storage concept with a 1TB/sec ingest speed. It suggest to me that object ingest and placement in an object store is slower than file ingest and placement in a file store. Obviously an object has to be hashed and then its location in the object namespace calculated and the data then written - which takes longer than tacking a file on to the end of a file store and updating a folder listing.

X-IO's CADP

X-IO has its Continuous Adaptive Data Placement technology (CADP) to identify sub-LUN hotspots on disk and move them into the SSD storage tier in its Hyper ISE and Hyper ISE 7 products. Hybrid flash array vendors can do tiering as well. IMHO the smaller the chunk of data the quicker it is to move it across storage tiers and/or into cache but the more data activity tracking there has to be.

Compellent with its move to a 64-bit O/S from its 32-bit forebear seemed to be positioned to track access levels on small chunks of data but hasn't announced any increase in tiering speed or granularity.

VSP does more than million IOPS

An HDS spokesperson sent this information to me:

>>>>>>>>>>>>>>>>

One thing I noticed though in your article that I was hoping could be corrected is the following statement: “Virtual machine density with a VSP can be doubled. HDS says an all-flash EMC VMX 40000 does 810,000 IOPS, about 20,000 less than a flash accelerated VSP.”

Our VSP with flash acceleration achieves more than 1 million IOPS, so I think the 20,000 figure above should be 200,000.

>>>>>>>>>>>>>>>>>

We should envisage a flash-accelerated VSP running at 1.2 million IOPS according to this statement.

Chris.

Glacier re-writes the rules

Amazon's Glacier (http://www.theregister.co.uk/2012/08/21/amazon_glacier_objjects/) seems to be a combination of object storage and spun down disks. It uses five data centres and will probably quite rapidly become the largest object storage implementation in the world. That will be a significant vote of confidence into object storage I think.

Chris.

Is Object storage really appropriate for 100+ PB stores?

Object storage vendors say their technology is ideal for storing billions of files and hundreds of petabytes of data in a single namespace across hundreds of sites. Is it?

CERN's LHC experiment stores its great mass of data on tape. Why not in an object store? Too expensive maybe? What is the real message about object storage if the billions of files/hundreds of petabytes use case doesn't actually exist?

Fusion-io ION server-into-flash SAN

With reference to the Fusion-io ION story about turning servers into all-flash SAN arrays, the company's CEO Davd Flynn sent me this about cost/GB:

" ... one point that my have been lost, Fusion-io sources raw NAND chips, while EMC, Pure, Nimbus and others are buying SSD's from 3rd parties. Enterprise SSD's can cost 2x what RAW NAND does - and then these other companies have to put their markup on it. Fusion-io has effectively eliminated one layer of margin mark-up by the time the customer gets the solution.

Then there's the fact that Fusion-io doesn't have to mark up the box that goes around the flash by 3x to make margin on it - since the customer sources that directly themselves. Open platform means fewer mouths to feed in the value chain - means lower cost. By being vertically integrated from raw flash to software defined storage solution we change the underlying cost structure...

Net net is that even without dedupe ION is likely to cost customers less per GB."

Our story said that Pure Storage and the other all-flash array startups typically use deduplication and compression to increase the effective capacity beyond the raw flash capacity and so decrease their cost/GB.

So ... is David right; would a customer pay less for a, say, 15TB all-flash ION SAn using a server from Dell or HP fitted with ioMemory flash cards, than for a 15TB all-flash array from Kaminario, Pure Storage or some other vendor?

Seagate didn't announce it was buying OCZ yesterday, as had been expected/hoped/rumoured. Its CFO did say Seagate could buy an enterprise SSD supplier with a significant share of the market, according to a Reuters report. So the premise that Seagate should buy an SSD company is weakened but still intact.

Seagate and OCZ

Should Seagate buy OCZ and get itself an enterprise and consumer SSD and PCIe flash hustler? There's rumours of Seagate bidding a billion or mire for OCZ with roughly half in cash and half in stock. Micron might be in there with a chip supply deal.

Is this good for Seagate?

Should WD/HGST do more in the flash business?

Phase Change Memory

A Micron Phase Change Memory product has just been announced. It's rated at 400MB/sec and we guesstimate it can do 100,000 IOPS. A PCIe card with ten of these products on it could do 1 million IOPS. Is this fast enough to knock out NAND?

Discussing TLC and write amplification

How about this thought: "50 full writes/cycles per day for 5 years does not necessarily translate to 89,000 P/E cycles. Please place close attention to the user/flash capacity ratio difference between this drive and the regular Optimis and Ultra. You will see that this ratio is reduced while the number of writes/day increases. This is because the WA is reduced as over-provisioning increases. I calculated the WA for the regular Optimus drive to be 2.58, while Ultra is 1.38 and Ultra+ is 1.03. Over-provisioning ((flash capacity-user capacity)/user capacity) is 28%, 71%, and 156%, respectively. The endurance required in all cases is about 37k cycles. User capacities start at 200GB, 150GB, and 100GB, for Optimis, Ultra and Ultra+, respectively."