|A quick look at some major storage standards|
What follows is a look at four storage-related standards: the Fabric Application Interface Standard (FAIS), Storage Management Interface Standard (SMI-S), RAID Data Disk Format (DDF) and iSCSI. The standards are presented in various stages of family unity, presented in rough order from the youngest to the oldest.
"There are always differences of opinion on any major standard," says Roger Reich, a member of the technical working group (TWG) responsible for SMI-S. "There are plenty of disagreements; that's part of the process."
All these disagreements can take a while to work out. For example, there were 20 distinct revisions of the iSCSI protocol before it got to a point where the entire membership of the Internet Engineering Task Force (IETF)--the guiding organization behind iSCSI--voted on it in April 2004. After four-plus years in the making, iSCSI is now officially considered a "proposed standard," meaning that the first version of the protocol is done. There will be additions and modifications in future versions, but the basics are set.
Generally, there are several pieces of any given standard. There's a core specification for iSCSI, for example, but there are also subsections related to security, naming, internationalization and other issues. Different members of subworking groups are generally responsible for those.
Then there's the matter of implementation. Many of today's standards require ensuring all of the resulting wares really do play nicely together, particularly because vendors are trying to get an early jump on incorporating them into products. As Gartner Inc. analyst Ken Dulaney explains, a standard doesn't necessarily "mean much to the end user until it is certified by an independent party. Without a certification body, there's liberty to implement it in many ways."
Of the standards profiled here, only SMI-S includes a separate product-certification effort to date, with plans to do so for RAID DDF-enabled products soon.
|Sleeper standard: NDMP|
FAIS is a means of standardizing the way that management and other types of storage-related software can talk to intelligent switches. For the immediate future, this standard is the most important to the vendor community. It will allow software suppliers like Veritas Software Corp. to create only one version of any given application that will then be able to run on different storage fabrics and across most of the major manufacturers' switches.
Ultimately, industry watchers expect that FAIS will affect users by resulting in more and different types of smart-switch applications and the ability to mix and match software from different vendors. But that's a couple of years away. FAIS is, at its heart, an application programming interface. So the people working on the standard are spending lots of time discussing how best to represent certain features. These are things that are implemented across different vendors' smart switches, such as programming calls and ways to boost performance by, for example, separating the control path from the data path.
Originally, FAIS started out as a proprietary Brocade Communications Systems protocol called XPath. According to Zulfiqar Qazilbash, co-founder and CSO at iVivity Inc. and a member of the FAIS TWG, the standard hasn't changed all that much since Brocade first offered it to a small group of interested vendors in early 2003.
The group is now "concentrating on the core features, and we've made a lot of headway in terms of what will be in version 1.0," Qazilbash says. "We're at a point where we've gotten 90% consensus and are discussing the last 10%." Both he and Brocade's chief technology officer Jay Kidd expect the first FAIS version to be pretty much agreed on by the end of this year. Then the various votes, feedback loops and other procedural matters should take until mid-2005. By the end of 2005, the first version of FAIS should be a done deal, participants say.
Of the standards reviewed here, SMI-S and iSCSI arguably have the most potential for the most long-term, direct impact on users. Several large companies have already included SMI-S as one of their requirements for storage contracts currently out for bid, according to two SMI-S spokesmen.
The main reason for this demand: SMI-S addresses a huge user need to more effectively manage different vendors' switches, RAID arrays and the like, and to share information among the different management packages. The way things work now, storage gear is often managed with proprietary software that tends not to work well with other vendors' software. So it's next to impossible to get information about how well your Brocade switch is performing, for instance, from your EMC Corp. Symmetrix-based storage area network (SAN).
SMI-S aims to solve that problem by extending some older standards, namely the Common Information Model and Web-Based Enterprise Management, both developed originally by the Distributed Management Task Force. SMI-S takes those as the basic transport mechanism and then adds an object model, an automated means of discovering new SAN components as they're added, resource locking and other features.
So far, 14 vendors--including AppIQ Inc., Dell Inc., Computer Network Technology Corp., Hewlett-Packard Co., Hitachi, IBM Corp., Network Appliance, Softek Storage Solutions Corp. and Sun Microsystems--have passed the first version of the certification test to prove their SMI-S products successfully interoperate. The Storage Network Industry Association (SNIA) runs its interoperability lab out of Colorado Springs and is currently preparing the fifth test bed in as many years.
Version 1.0 of SMI-S has been available for more than a year, and revisions are coming in the next six months to a year, says Phil Kemp, who works with SNIA's Storage Management Forum and whose day job involves product marketing at HP's storage software group. Enhancements on tap for version 1.1 include performance management and measurement and the addition of tape libraries to the management mix. There will also "be an attempt to do logical file systems on hosts," he says. After that, the SMI-S roadmap calls for adding service level agreements and policy management, perhaps in version 1.2 of the standard.
|Advanced Switch Interconnect|
RAID DDF is designed to allow users to swap their vendor's RAID device for another to do what's called an inplace migration. "If a customer has a RAID set on a server that's direct-attached, and then wants to move that to an internal enclosure, you can migrate the RAID set," explains Larry Lamers, director of technology and standards at Adaptec Inc. and secretary of the DDF TWG. The other key benefit is if users run into a failure with a specific RAID controller, they can replace it with a different vendor's controller, in place of the original.
Lamers says that the TWG is "pretty much in agreement" over what the standard should include. "This one was pretty straightforward." Interoperability testing should take place the first half of next year, with products on tap for late 2005, he says. The group is currently working on profiles that define interoperability points.
The standard currently includes two basic elements. One is a data structure, to define which level of RAID has been implemented. This lets every disk know which other disks are part of the RAID configuration. The second piece is the actual layout--stripe size, stripe depth, etc. Today the metadata that describes how each vendor represents its RAID device is supplier-specific, and the goal is to allow RAID disks to recognize each other.
Although the DDF's list of supporters and participants isn't public yet, Lamers says, "I can't think of any RAID vendor with any significant market share that's not involved in this. Some came in late, but everybody pretty much joined the parade."
So far there have been two prototype implementations of DDF, says Ramamurthy Krithivas, an architect with Intel Corp.'s storage components division and a member of the TWG. He says he'd like to see the next version of the DDF incorporate a basic reporting mechanism for RAID 6 that wasn't included in the first version of DDF "because of all the complexities" involved. "I'm talking about the configuration information, not the layout, because the layout is where vendors add value."
Most industry watchers expect iSCSI to help users connect second- and third-tier servers into SANs that have been previously reserved for only critical applications. iSCSI will also help smaller firms implement SANs for the first time over their existing network infrastructure without needing to install Fibre Channel equipment.
In the lingo of the IETF--the organization shepherding this protocol--iSCSI is a "proposed" standard. That means "it's been agreed upon by an IETF working group, it's complete and secure and has its bases covered," says Mark Bakke, who is a technology leader with Cisco's Storage Technology Group and also sits on the iSCSI TWG.
It's at this point that vendors go out and implement, and "we find out how well all the products work together," Bakke says.
The next level up is considered a "draft" standard, and its function is to clean up any bugs, remove any optional pieces that nobody implemented and other housekeeping tasks. "There's no big rush to do it and there are no current plans to advance iSCSI to a draft standard, but I expect we will because the standard is so popular," Bakke explains.
He says there are many products with iSCSI included because the core functions of the protocol "haven't changed in two years." From an operational standpoint, the products based on iSCSI have remained constant for the past several revisions.
That said, there are some new things to watch for. One is a layer that defines how iSCSI can run over InfiniBand or over Remote Direct Memory Access, so the IP network can handle the same kind of high-bandwidth, low-latency applications that traditionally have required more expensive alternatives.
Other pieces will relate to various management information bases (MIBs)--in other words, defining modules to plug into the Simple Network Management Protocol and that gather network-related statistics about and from different types of gear. Those MIBs will come out at the end of this year, Bakke says. And there will be a mechanism to boot diskless devices over iSCSI that Cisco, for one, is already supporting but isn't yet official.
Taken as a group, it's still an open question if these standards will make storage managers' jobs easier anytime soon or simply serve to complicate what can be an already convoluted situation. Then again, that may be an argument better left for another day.