Published: 10 Oct 2006
Developing standards in the storage industry is like achieving consensus at the United Nations--it takes a while. The Extensible Access Method (XAM) Interface, which would define a single access method for archiving devices like EMC's Centera and Hewlett-Packard's (HP) Reference Information Storage System (RISS), is no exception.
On the surface, XAM appears to have everyone's support. EMC, HP, Hitachi Data Systems (HDS), IBM and Sun Microsystems are all in the same camp. All agree that a single API--instead of proprietary APIs developed by each vendor--will grow the market much faster. Application providers will only have to write to one API and, more importantly, users will be able to move data from one archival device to another (from Centera to RISS or vice versa) without much hassle. Today, users would have to convert all of the data into the format of the new device, changing addresses and hashes and potentially compromising the integrity of the data in the process.
Carl Follstad, manager of data management services at the University of Minnesota in Minneapolis, is evaluating digital archiving systems, but says he's "hesitating to do anything because of the lock-in problem." The University of Minnesota is an all-EMC shop with a 300TB SAN and is reluctant to buy any more disk storage.
"I like Centera because of its longevity in the industry, but I want a software-only solution," says Follstad. He notes that the university is ultimately looking at converting its 300-year-old library to a digital archive.
"I can't be spinning this off lease every three years," he says. "I need something that I can seamlessly migrate over a long period of time."
The question is whether vendors can agree on the best format for the standard. Three approaches are up for discussion: a client-side API or driver, a pure protocol-only approach and a file-system version.
Leading the client-side driver camp is EMC; its existing Centera API is also a client-side driver. Mike Kilian, CTO at EMC's Centera Division, says it's more flexible for supporting a diversity of devices, some that may not even be IP-attached, like FICON systems. With the driver approach, standardization is done at the API level and individual drivers (called Vendor Interface Modules or VIMs) translate the API functions to any type of network or transport protocol. XAM doesn't enforce which protocols vendors must use to be compliant. Standardizing an API is easier and faster than standardizing a protocol, says EMC.
But the pure protocol camp, lead by Archivas and its partner HDS, argues that this isn't the best approach for the industry.
"Client-side code has to go through cycles of qualification and verification that elongates the rollout of new functionality and introduces complexity to users," says Rob Mason, VP of engineering at Archivas. Mason is trying to steer the group toward a pure protocol-only approach like the Simple Object Access Protocol (SOAP) or JSR-170, which he believes to be a "cleaner implementation" and easier to update.
In the third corner is Network Appliance, which hopes the XAM interface will also support NFS and CIFS, in keeping with its file systems.
Zoran Cakeljic, chief technologist in EMC's Centera group, believes the polarity of interests within the XAM community can all be represented. "The protocol camp can have a standard protocol in the future ... and we've found an elegant way to map file systems onto XAM, but the driver camp will hit the road sooner," he says, in the interests of getting something to market.
Cakeljic hopes the Storage Networking Industry Association (SNIA) will design a standard protocol and produce a standard VIM that implements that protocol. This standard VIM would coexist with vendor-supplied VIMs within the same API framework, and any storage vendor wishing to speak the standard protocol could use it. At this point, vendors will have a choice between maintaining their own protocols and supplying their own VIMs, or subscribing to the standard protocol and not having to supply any vendor VIMs.
"The latter is probably technically the better place to be, but it is also further away in the future," notes Cakeljic.
How this two-step process toward a standard protocol will impact users isn't exactly clear, but at least it's moving in the right direction. The XAM group expects a draft of the specification to be complete by year's end; interoperability demos will be presented at Storage Networking World next spring, followed by a certification program the following fall.