Analysis

posted on 28 July 2008 09:52

In its earnings call (transcript here) deduplication system supplier Data Domain (DD) raised its guidance and gave more visibility into its product roadmap

Outlook

CFO Mike Scarpelli said:  "... we are raising our previously established guidance. For the third quarter, we expect revenue in the range of 67 million to 69 million. Product revenue will comprise approximately 56 million to 58 million of the total, while support and services should generate roughly 11 million. For the fiscal year 2008, we are increasing our revenue guidance to 250 million to 255 million, up from our previous guidance of 225 million to 235 million. We will revisit this guidance at the end of Q3."

"On the cost side, we will continue to add headcount at a similar rate in Q3 to the second quarter particularly within the sales organization. Additionally, we will be aggressively staffing our Research Triangle Park, North Carolina facility in selected functional areas such as R&D and support. We will also continue to invest heavily in our direct marketing campaigns, both domestically and internationally, to drive mind share and customer lead generation, which we feel is a key component to support our revenue growth."

"As such we anticipate a third quarter 2008 GAAP operating profit to range from breakeven to 500,000 which includes stock-based compensation of 6.6 million to 6.7 million. Moving to earnings per share, we are forecasting earnings per share on a GAAP basis to range from $0.01 to $0.02 per diluted share."

Competition

CEO Frank Slootman said: "... the competitive lineup really hasn't changed that much. It's still EMC probably in half or better of the transactions that we do contest. And after that it's a mixture of NetApp and Quantum. We do feel that Quantum after the OEM relationship with EMC might have backed off a little bit from their own direct push, but that's sure anecdotal perception on our part. NetApp being very strong in their own installed base but not that much beyond that. After that it becomes a longer list of people that don't have a bigger presence. In terms of HP and IBM, we never saw those people that much and we don't expect them to see them that much going forward. They sort of have their own customers, they own channels, and as I said before, they sort of live on their own continent, they're sort of not that present in the broader marketplace where EMC and ourselves are very, very active."

Customer use case

Slootman said: "... the vast majority of what we're dealing with is massive replacement of tape infrastructure. We do have some percentage of people that have sort of struggled along with what we call first generation disc backup and they're really not getting any or are still making tapes being pretty unhappy. And mostly those are the VTL category of customers. So, we're seeing more of that as well. But for the most part, it is tape replacement."

Roadmap

Slootman said: "Over the last four years, between 2004 and 2008, actually five years, our systems have increased in throughput by 10x and in terms of capacity by about 30x, and we're guiding our customers to continue to expect that trajectory going forward. And it's driven – one of the unique things about the date if the main architecture is that we are one big bet on the Intel roadmap, not on Seagate, so we're following Intel roadmap very closely, and that gives us the opportunities to expand our systems at the rate that we have. So, you can expect single nodes from us, and we’ve also said in a number of different forums and venues that we’re also working on releasing a cluster configuration of our 690 systems and that will take the performance and scale gain to a completely different level."

Let's see than. Currently we have quad core Xeon CPUs from Intel. What we can expect are a 6-core Dunnington processor by the end of this year, if not before, and then the 2 - 8-core Nehalem (pictured left) CPU architecture with 4-16 threads, that is two per core. It will span the desktop-workstation-server range and we might expect server versions in 2009.

If DD is keen to keep a performance edge then it will use Dunnington as a short-term processor in its products and then progress to Nehalem. Nehalem will require more software recoding to take advantage of having two threads per core.

So we might presume there will be a Dunnington upgrade or addition for all DD's controller systems: 565; 580; and 690, in the last quarter of this year or early 2009. In the second half of 2009 or early 2010 we will see Nehalem upgrades or additions with a range topping 8-core product, possibly called the DD 7XX something or other.

Cluster

A clustered 690, effectively replacing the 690g aggregation, would need an interconnect and system software to be cognizant of operating in a cluster with cluster node state information exchange, load-balancing and some form of failover. If there is a cluster wide de-dupe table then there needs to be a lightning fast cluster interconnect to keep the inline de-dupe working. That would suggest InfiniBand (20 or 40Gbit/s) or 10GbE. We might also consider that no interconect could keep up with real-time additions to a cluster node's de-dupe map and envisage DD adding post-processing de-dupe options to take care of cross-cluster node deduping possibilities.

SSD

A third roadmap option centres on solid state drives (SSD). In answer to a question from Wachovia analyst Aaron Rakers, Frank Slootman said: "... this is an area of enquiry and research for us. It's quite interesting at a number of levels for us, but we're also not in a position to make statements on that technology at this point."

Our thinking is that a possibility is for the dedupe map to reside in the controller in a very large SSD cache. A Samsung server-focussed SSD might then be a point of interest but we imagine that its write endurance would be a critical measure for DD as mulltiple CPU cores, and multiple threads, would pummel the SSD resource.

A second option is to add SSD technology as a fast tier zero in DD's existing storage arrays. To the extent that DD controllers can front-end third-party arrays then that will come to pass as third-parties provide SSD-enhanced arrays in, guesstimates say, the second half of 2009 onwards.

It is entirely possible that DD's software engineering resource has simply too much to cope with. The imminent arrival of Dunnington with six cores gives its work to do. The arrival of clustering gives it a substantial addition of extra work. There might simply be no bandwidth left for working on how to use SSD-enhanced controllers.

^{[Chris Mellor.]}