Columns
A New Kind of Networked Storage
A new storage solution called Network Unified Storage offers a compelling idea.
In the context of storage technology, "networked storage" poses something of a Zen riddle.
On one hand, virtually all storage is networked, because local and wide area networks are used to provide data access to applications and users. On the other hand, no storage platform that's offered today is really networked storage. Rather, there have been only types of storage: server-attached, thin server-attached and switched server-attached.
Now, a new kind of storage recently presented to me by a newcomer company to the storage arena makes it important to understand the difference. Let me explain.
Industry efforts to discriminate server-attached arrays and internal disk storage from so-called "server captive" storage (NAS and SAN) is dubious and self-serving. Network attached storage, or NAS, is actually a kernel operating system—a "thin server"—wedded to a storage array. Strictly speaking, NAS is simply server-attached storage, but without the extra overhead associated with a general-purpose operating system. The NAS thin server kernel provides a subset of general-purpose OS functions required for IP processing, network file system emulation and storage I/O.
Similarly, storage area networks, or SANs, are simply switched server-attached storage platforms. In a Fibre Channel SAN, a switch is placed between servers and storage devices in order to make and break point-to-point connections at high speed. In point of fact, using Fibre Channel and Storage Area Network in the same breath is oxymoronic. Because Fibre Channel is not a network protocol, anything built with it is not a network. The most you can create from the high speed serial interconnect called Fibre Channel is a fabric—a bunch of point-to-point, server-attached storage platforms.
Neither NAS or SAN are, strictly speaking, networked storage. Both technologies perpetuate the architecture of server-attached storage. They simply add a few wrinkles that solve certain application problems while introducing others.
For truly networked storage platforms to appear, operating system vendors, hardware vendors, and system architects—not to mention storage consumers—need to be prepared to break with traditional thinking. We're compelled to do this by certain realities of storage in our enterprises—rapidly increasing data and rapidly approaching limits in hard disk storage densities. But because of inertia, market fears and budgetary constraints, you may be reluctant to make this break.
Networked Storage: Points to Consider 1. Most of the new networked storage technologies use proprietary block storage protocols. (This goes for the latest implementations of iSCSI announced by industry stalwarts in February, because these use reference versions of the iSCSI protocol rather than the current protocol in development at the IP Storage Working Group of the Internet Engineering Task Force.) Before you buy, be sure to ask this: What does deploying a non-standard protocol mean going forward? Try to avoid vendor lock-in. 2. Ask about load balancing. If you'll be sharing networked storage platforms, ask how transactions will be handled in a load-balanced deployment scenario. Fail-over capabilities are desirable, but not if they confuse applications or storage devices. 3. Ensure application fit. Make sure the networked storage platform will support the application for which the storage is being deployed. 4. Is it open? Determine whether you'll need to buy all storage devices from the vendor or if the platform allows you to include heterogeneous platforms in the future. 5. Be bold. Networked storage is exciting and new. Early adopters are true visionaries who are willing to break with time-honored concepts of storage to bring about an entirely new storage infrastructure. Expect early challenges, but emphasize management and monitoring capabilities in product acquisitions. —J.W.T. | |
Storage vendors are also reluctant. Hardcore server-attached array providers don't want a real networked storage technology to appear until they're ready to sell the platforms themselves and to ensure that only their own equipment can be used. That preserves their bottom line.
On the other hand, newcomers to storage—Left Hand Networks of Boulder, Colo. for one—lack the bunker mentality of the old guard players. They're not concerned about the best way to protect existing market share (since they haven't any), and they're generally unafraid to challenge conventional thinking on storage technology.
Left Hand's CEO, Bill Chambers, demonstrates palpable excitement when he talks about cutting the umbilical cord of server-attachment in storage technology. He and his management team—most of whom have long pedigrees as industry insiders—are advancing the proposition that it's time for storage to "grow up" and to become the standalone networked infrastructure that it needs to become if companies are to realize their data intensive and cost-sensitive business process objectives.
Left Hand, whose name derives (depending on who you ask) from the southpaw orientation of its senior management, the name of the Rocky Mountain canyon visible from the back door of the headquarters building, or the sign over the brewery where Chambers and company first defined their new storage concept, is on to something important. They call their solution Network Unified Storage (NUS) in order to discriminate it from current NAS or SAN or other carryovers from the server-attached world. Whether or not you like another acronym (I don't), you have to like what it stands for.
Left Hand's NUS leverages both open and proprietary protocols to perform both block- and file-based storage I/O in a network. Network file system protocols such as the Common Internet File System (CIFS) and Network File System (NFS), and eventually the network block access protocol iSCSI, are supported, but the company also offers its own Advanced Ethernet Block Storage (AEBS) protocol for very efficient raw block storage today. AEBS sends raw blocks of data across a network (Gigabit Ethernet is currently the preferred medium) to an intelligent, highly scalable, storage cluster array, where they are executed.
Simple message passing avoids many of the troublesome concerns of handling chatty SCSI command traffic over IP stack processing networks while delivering the performance required by block-intensive high performance databases.
Additional innovative technology is used on the backend of Left Hand's NUS platform to provide concurrent support for block and file storage and to enable the continuous scaling of storage capacity with full RAID 10 mirroring and striping across a theoretically limitless number of storage arrays, called modules.
Each Network Storage Module has an Intel Pentium family processor and offers four IDE/ATA drives. Superb engineering has gone into the instrumentation of each module to facilitate real-time monitoring and management, and some very effective design measures—worthy of emulation by all storage manufacturers—have been taken to reduce power consumption requirements and heat generation properties of the modules.
As you would expect from an enterprise storage technology, the NUS modules also feature redundant everything: Eight cooling fans, two power supplies, duplicate internal and external network interface adapters, etc.
Currently, Left Hand provides NUS modules in increments priced competitively with traditional storage solutions in the market: $15,000 for a 160GB module, $25,000 for a 320GB unit, and $35,000 for a 480GB system. The modules themselves are clustered through parallel Gigabit Ethernet interconnects and a minimum of two modules are typically purchased at one time to support cluster fail-over.
Left Hand's offering is the first of what I hope will be a flood of true networked storage products. While NUS is an exciting new development in the market, and one that has interested several venture funding organizations in Left Hand, the technology's value has yet to be demonstrated. However, NUS has been adopted by several large healthcare and Internet companies, mostly in the Denver area, and more sales are in the works, according to Chambers, who says that he is ramping up his direct and channel sales operations. Clearly, prospective customers will need substantial face time to help them understand what's different about Left Hand's approach.
My advice to readers is to keep an open mind when Left Hand (or its eventual competitors) comes knocking. Real networked storage is truly a different beast from what established vendors have been selling since the mid-1990s. When loud contrarian views begin to be voiced by traditional storage vendors, as they inevitably will be as this new technology garners more market share, keep in mind that noise levels tend to increase when marketers have little in the way of convincing counter-arguments.