Bigger Can Be Baffling
EMC's latest mega-storage product leaves our storage analyst baffled
As I read the announcements about EMC’s latest entry in the “enterprise-class” storage array market last week, the DMX 3, with its initial 288 Terabyte storage capacity on 960 disk drives increasing to a Petabyte on 2,000 drives by the end of 2006, I scratched my head in wonder. Why would the company pursue a course of Bigger Iron at a time when the market for such wares is actually shrinking?
How many people actually need such capacity, I wondered: a few oil companies, maybe the largest of the big banks and credit card processors, a couple of government agencies, and let’s not forget the largest telcos. Beyond this handful of vertical niche players, I couldn’t think of anyone else who needed such an array.
Indeed, the market that DMX 3 targets is already saturated by both HDS with its TagmaStore offering and IBM with its current and next generation Shark arrays. It was my impression that DMX 3 is just the latest entry into an already over-served market in which customers, according to Wall Street, aren't inclined to buy a lot more. Try as I might, I could make no sense of the Hopkinton move, except perhaps that it was an effort to keep up with the Joneses in Santa Clara and Armonk.
To provide a sense of scale about the meaning of a Petabyte, an EMC spokesperson mentioned that DMX 3 could store up to 250 million MP3 music files. I mused that maybe EMC would position DMX 3 as a kind of platinum iPod accessory: the perfect Christmas item for the typical high-end shopper with a spare $250K (entry-level pricing on the array) burning a hole in his pocket and no original thoughts about how to spend it.
Suddenly, déjà vu washed over me. DMX 3, I realized, is a lot like the M1 Abrams tank when it first came out in 1978. Some of us are old enough to remember the stories: U.S. taxpayers not only needed to buy the mobile artillery piece, at X-million dollars a copy, we also needed to buy the earthmover to flatten out the terrain in front of the armored vehicle and the fuel tanker to follow behind to continuously top off the gas tank every few kilometers.
Industry insiders suggest that EMC is effectively telling customers, who in most cases have just purchased and installed DMX 2, that their existing storage, with its non-mirrored cache, is vulnerable (something they previously denied) and that they need to spend more money on bigger iron, plus buy the Invista virtualization product (required for promised throughput improvements), plus whatever else they can think to add, to get to the promised land. (EMC denies that non-mirrored memory caches are vulnerable. Spokespersons claim that "DMX 2 is selling as well as ever" with its non mirrored cache and the only reason why mirrored memory was added in DMX 3 is that "the chips became available for doing it." So, with this spin of the facts, it would seem that customers are paying more than they need to for features that are unnecessary.)
The parallel goes even further: the real value of the M1 Tank to its manufacturer, Chrysler Defense (later purchased by General Dynamics), was not the money received from the initial product sale but the spare parts contract that followed. In the same vein, with 2,000 disk drives and an estimated annual failure rate of about 3.7 percent, EMC stands to make a ton of money just replacing drives at a statistical pace of about seven per month. (This fact becomes even more important when you realize that all drives must be sourced from EMC - see below.)
The vendor has apparently anticipated this and simplified replacement by adding “hot swap” capabilities. That way, users can self-service the arrays rather than calling in a tech every time a drive goes south.
Of course, I’ll bet that the revenue lost from those service calls is made up in component costs. If things remain as they are today, you will need to buy the drives from EMC directly, and at a mark-up over what you would pay for the same drive on the street or if purchased directly from its manufacturer.
Let’s not forget that the $250K ante on the initial model does not include software. If it did, I’ll bet someone would have mentioned this important point. (No one did.)
So, add in the annual cost of licenses for Symmetrix Remote Data Facility (EMC’s external mirroring product that works best only if you have EMC products on both sides of the mirror), TimeFinder (EMC’s wasteful on-array point-in-time mirror-splitting product), Enterprise Control Center (Hopkinton’s handy-but-flawed storage management package), and whatever other software they can mount on the box, and, to paraphrase the quip from the late Senator Everett Dirksen, “Pretty soon, you’re talking about real money.”
Also, let’s not forget the cost of attachment. This box is initially targeted at mainframe shops with its FICON interface, but the Hopkinton gang says that iSCSI and Fibre Channel connectivity options are also available (or shortly will be) to meet the needs of the distributed world. You will be relieved to know that soon you can connect the $1 million-plus Big Iron array behind a $150 Linux server if you want, provided, of course, that it is already connected to a $3000-per-port FC fabric switch. Fibre Channel is a huge cost accelerator, boosting the per GB cost for storing data from about $80 per GB for enterprise disk to $180 per GB when that disk is connected to a SAN.
Some analysts might caution that it might be best not to consider the price tag as a predictor of the success of DMX 3. EMC is known for making relationship sales in the front office rather than technical sales in the back office.
Put bluntly, hardly anyone buys EMC because of the technical superiority of its products. That’s one reason it struck me as odd to hear the company emphasizing “linear scalability,” “double the backplane bandwidth and processing power,” and “fully mirrored global memory directors based on dual data rate SDRam technologies.” That’s a lot of geek speak that, even if it actually meant something, probably wouldn’t sell a single box.
EMC markets to “Flashing 12s” pure and simple: C-level executives who can’t program their VCR clocks (which subsequently flash 12:00 forever). More often than not, they sell the CFO after they have been shown the door by the IT professionals who have reviewed their products and found them lacking.
Even in strictly technical terms, I found myself questioning the value proposition of the new array. The product sounds less like a state-of-the-art-advancement-in-storage-technology than a much-needed-and-long-overdue (and still technically deficit) upgrade to a dinosaur platform. Where was the support for multi-tenancy provided by HDS with its embedded virtualization and logical partitioning? Where was the deconstructionalist-software-with-modular-hardware-support-for-upwards-AND-downward-scaling of IBM? Faster backplanes and failover caches do not a paradigm shift in storage make.
The EMC pitch emphasized the ability of DMX 3 to support—in the same array and together with their normal ranks of overpriced, FC drives—new, lower-cost, and more capacious Fibre Channel (LC-FC) disk drives, which are the rough equivalent of “F-ATA drives” (high capacity ATA disk with a FC interface) in Hewlett Packard-speak. This design choice, they said, will enable them to build “multi-tiered storage” inside a storage cabinet “thereby facilitating Information Lifecycle Management.”
Apparently, existing standards like S-ATA (Serial ATA), S-ATA II, and Serial Attached SCSI (SAS), were just not up to the job in Hopkinton’s view—not that there is anything wrong with SATA or SAS. It’s just that using them wouldn’t provide EMC sufficient differentiation from the pack. At first blush, their reasoning in this regard seems to be on a par with that of the cliquey kids we all knew in high school, those who described everyone else as their inferiors to stress their differences.
However, a follow-up conversation with EMC suggested another view. According to spokespersons at the company, LC-FC drives were selected because they were compatible with EMC's existing controller and did not require a redesign of the array. They went on to note that "nobody else has implemented SAS/SATA into products designed for the high end market." True enough.
"Customers love having the lower cost disk inside the array," they said. But when I queried whether consumers would be able to purchase off-the-shelf LC-FC drives for use with the array, they waffled. EMC drives contain special microcode to work with the array, they reported, and this is not made available to consumers who might prefer to source their own components. The reason: quality assurance. Asked if there was some concern about the quality of drives as guaranteed by their sole manufacturer, Seagate, their response was, "Just say we want to err on the cautious side and leave it at that." So, expect to pay up to three times the street price for the Lower Cost FC drive you buy with the array in any case.
Doing “multi-tiered storage inside the array” was a particularly interesting bit of double-speak. Multi-tiered storage is not the result of hardware but of software. Without an application such as Hierarchical Storage Management, disk is just disk, whether it is S-ATA or F-ATA, located inside an array cabinet, in a fabric, or in a server. EMC knows this well, but is once again dumbing down the vocabulary in the interest of marketing.
Perhaps the biggest head-scratcher of the whole DMX 3 pitch is wrapping this double speak in the flag of ILM. A key goal of ILM is to reduce the need for more storage capacity by managing data better. This is accomplished by leveraging effective data-management processes to cull stale and contraband data -- which comprises up to 80 percent of the bits eating up precious capacity in most storage arrays today. By grooming data in a disciplined manner, ILM forestalls the need to buy more capacity, which is clearly not what EMC has in mind as it proffers the biggest Big Iron on the block.
In positioning DMX 3 as an ILM silver bullet, the vendor has shown its true colors. Despite the small fortune in marketing that Hopkinton has spent to put Information Lifecycle Management on the front pages of every business and industry publication, in the keynote position of every industry conference in the world, and in the reading pile of every corporate executive washroom of the Global 2000, they are really just about selling boxes of disk drives after all.
Centera, their platform for long-term data retention (but only if you buy only Centera), was just a teaser. DMX 3 is the ultimate expression of the EMC vision: thinking inside the box and owning the customer by owning his data.
That’s my view, but I’d like to hear yours. If you are an EMC customer and you are bound by your warranty not to speak publicly about the performance you receive from your gear, I will keep your information on the QT. Send your reviews to firstname.lastname@example.org.