In-Depth
Heterogeneous Snapshots Have Arrived
Will customers be able to establish real storage tiers and perform migrations of data per policy to enable 90 percent capacity allocation efficiency with a new arrival on the array horizon … and say goodbye to proprietary products?
The past week has been filled by vendor messaging about improvements in array technologies. Hitachi Data Systems (or an observer) posted a funny commercial on YouTube depicting storage arrays as brain-dead zombies guided by an evil consultant stating that virtualization should be done off-frame, in the network. Fortunately, Mr. T (yes, that Mr. T of Rocky movie, A-Team, and cartoon fame) bursts through a flimsy wall and explains that the Universal Storage Platform (USP, called TagmaStor by HDS) does virtualization the right way: on the storage array and as a function of a controller.
While I suppose that some people feel comfortable taking their storage architecture guidance from Mr. T, I prefer the message of vendors seeking to deconstruct proprietary, homogeneous, and monolithic storage infrastructure and who wish to see any-to-any, network-based architectures instead. One man who speaks eloquently to this point is Acopia Networks CEO Chris Lynch.
Lynch brazenly refers to the big storage players as a “storage cartel”—“an oligopoly that artificially maintains the prices on its products through homogeneous proprietary systems that stifle innovation.”
Lynch is more than just another “Peck’s Bad Boy” of the storage industry. He's head of a small company with an axe to grind with the brand names. Prior to taking the big chair at Acopia, he was Vvice president of worldwide content delivery networking sales, where he led Cisco to a leadership position in the content delivery market. Before that, he was vice president of worldwide sales, marketing and support at ArrowPoint Communications, and veep of sales for Lucent Networks. Lynch is a bona fide network person who believes that functionality built into arrays to prevent cross-platform management should be placed instead in the network as services that can be used by all storage devices.
He has put his money where his mouth is by developing a new capability, called heterogeneous snapshot, as a function of his switch platform. Put simply, you can do a snapshot of any storage array with his approach and place the resulting data image on any other array you might have installed in your shop.
As this column goes live, we will be the first to announce the product and I will be up in Lowell, MA validating the functionality described. That is how important this is. How, to use Lynch’s own words, “economically disruptive” it is to the fortunes of vendors who have made proprietary snapshots part of their storage product architecture in an effort to lock customers into their technology.
Lynch says that his technology is “technically complimentary to EMC and Network Appliance” but he also goes on to note that deploying his technology ahead of NAS heads and doing snapshots at the network layer generates a “300 percent cost savings” for customers in contrast to using the proprietary software offered by leading array and filer makers.
“We are network guys,” Lynch says, “not storage guys. Using our product enables every shop to become [at least] a two vendor shop—you can use any storage you want.”
Here’s how it works under the covers. The customer deploys an Acopia object-based switch that virtualizes back-end NAS arrays. The customer can then create snapshots in Acopia’s virtualized volumes, on-demand, using external agents and coordinated with applications. Snaps are policy-based and occur at pre-scheduled times as dictated by the customer.
The approach is integrated with Acopia’s replication services, a switch-based data mover that provides a robust approach for moving files around infrastructure. Snapshots are granular, with the product creating one filer snapshot in each filer volume hosting shares from an Acopia volume. Currently, snapshots in volumes backed by EMC and NTAP filers are supported; Lynch says that others will be added shortly.
An Acopia snapshot obeys all consistency requirements: the snapshot preserves “write ordering” within the Acopia volume (i.e., across multiple filer volumes and filers). The snapshot contains no “torn writes” (these occur when an unplanned event such as a power failure prevents the system from completely writing a block of data to disk), so, Lynch argues, “it satisfies all requirements for replication, backup, recovery, etc.”
What Lynch really wants to crow about is performance and efficiency. According to the energetic CEO, doing snapshots in the network on an Acopia Networks switch provides a 300-to-1 performance bump over doing snapshots the NetApp or EMC way, at the filer level. Moreover, using the other data management capabilities of the switch provides customers with an off-platform, highly manageable mechanism for doing real hierarchical storage management (HSM)—with the Acopia switch providing “Tier 0” functions.
“With our thin provisioning,” Lynch asserts, “customers can establish real storage tiers and perform migrations of data per policy to enable 90 percent capacity [allocation] efficiency [in a distributed environment].” If true, this would rival the allocation efficiency of DASD in a mainframe environment.
I must say that I am excited to travel to the frozen North from my base in Florida to see this in action. I will report what I find—especially if there is any vapor in the Acopia value proposition.
Bottom line: HDS marketing and creative folks may think that virtualization and similar services, including snapshots and data migration, belong on the array frame. Not everyone agrees.
One long-time industry insider, LeRoy Budnik, who is the managing partner ofKnowledge Transfer, told me recently that the view of storage controller functionality moving to the network “is appropriate, and supported by history.”
“In long standing Intel commentary, they suggest that technologies that move down the stack tend to succeed in the long term; hence, storage virtualization [and other storage services] moving toward the network is a successful pattern.”
In some respects, Budnik reminded me, leading vendors are already doing the enabling work, but not really admitting it. Although the HDS and EMC products have differently distributed architectures, what they are really building is a controller with embedded virtualization. The disk could easily be torn away, and the controller head could be sold by itself to front-end storage in a network—rather like the Acopia Networks offering.
However, Budnik adds, “the market is not ready for that tearing—yet. People like a cohesive package. Hence, they continue to pay a premium for the package to the benefit of the industry.”
He says that some vendors have suggested moving everything back to the server hardware/OS level, “After all, this is where things like RAID started as an application, later to be embedded in chips. This is good for developing new services but difficult for sustainable services, even those services that are now in CPU silicon. It goes against the model.”
In fairness, Acopia is currently front-ending only NAS and not block storage arrays installed into Fibre Channel fabrics. “Recent history has shown that there are difficulties coordinating a virtualization engine across multiple FC fabrics,” according to Budnik.
“We worked under the concept of dual fabrics as an availability mechanism, and frankly, people have made so many accidental mistakes, it has been necessary. The single fabric, yet partitioned, view using [Cisco Systems] VSAN has not caught on to the degree expected by its supporters. That technique had better potential for creating network-based [storage services].”
While Budnik’s words resonate on several points, they also fly in the face of trends we are seeing in the industry. Companies such as Acopia Networks and others (Zetera, DataCore Software, and SOFTEK, among others) keep popping out of the woodwork, laboring “around the edges” of myopic brand-name storage vendors to bring network-based alternatives to “everything-on-the-proprietary-array-controller” solutions of the big guys. There are signs that some of the big vendors are catching on.
On January 29, for example, IBM announced that it was buying SOFTEK, an independent and privately-held software company that we have discussed in this column. SOFTEK has been leveraging and building on a mainframe data migration utility developed originally at AMDAHL some years back, called TDMF. Improved and extended into the distributed computing world under the moniker “Softek Replicator,” the software provides the means to migrate block data across and between different vendor hardware, regardless of brand or model number, in an intelligent and reliable way.
It was predictable that IBM would brag that the acquisition provides a fast and dependable method for customers to drag their data off EMC frames and onto IBM arrays, supporting product choice. However, the technology was originally vetted with EMC to facilitate exactly the reverse path. (This is reminiscent of how AMDAHL—and later Fujitsu—framed the value of TDMF, as a mechanism for moving data off IBM DASD and onto alternative storage.)
The irony is inescapable: TDMF, which began life in the hands of IBM competitors, has now come home to Big Blue. The good news is that whichever vendor’s gear you choose to deploy, IBM’s newly acquired utility provides flexibility and choice in the storage gear you deploy.
My next column will be written from a hotel in Lowell, MA. Until then, feel free to drop a line to me at [email protected].