Out-of-the-Box Thinking From Zetera
Despite the tens of millions of dollars spent by vendor marketing departments since the late 1990s to evangelize network storage, it has never happened. Until now.
When you think of storage, you might imagine a nondescript chassis filled with disk drives from Western Digital, Seagate, Maxtor, Hitachi, et al, arranged in orderly ranks and cabled back to a fancy RAID controller. If you use brand-name storage, you may also envision a pretty cabinet and door of Italian design decorated with a splendid logo for your vendor of choice. (You pay a lot extra for that logo, by the way.)
Typically, there is a cable or two (or more for the bigger arrays) coming out of the array that connects to a server or to a switch (or two), providing a path for SCSI commands and data traffic. Another cable, usually Cat 5, or better Ethernet, also springs from the cabinet to connect to a LAN that provides the “control path” for the array—an obtuse way of saying management and security services.
That’s storage in a nutshell. Everybody is selling a box of disks, some with fancy controllers and redundant components, such as power supplies and fans.
Bottom line: every array is channel-attached, connected directly to a server. I know you have heard the terms network storage or network-attached storage or storage area networks. As my Italian relatives are fond of saying, “Forgetaboutit.”
NAS boxes operate a server OS, just like a file server, and their drive trays connect back to this server, so NAS is still server-attached configuration.
FC SANs are simply server-attached arrays whose data-path cabling just happens to pass through a simplistic switch that makes and breaks point-to-point connections at high speed. The introduction of a switch between server and storage device doesn’t make it a network by any accepted definition of the term, except for the definition used by marketing departments of FC switch makers. The FC switch serves the same basic function as a roundhouse in a toy train set—you know, that expensive piece that rotates and transfers toy trains from one set of tracks to another.
Despite the tens of millions of dollars spent by vendor marketing departments since the late 1990s to evangelize network storage, it has never happened. Even iSCSI, the operation of the channel storage architecture SCSI as an application across a true network (IP), isn’t the real deal when you get right down to it. It does bring data path and control path into the same wire, but it perpetuates the same initiator and target relationships that have existed since storage was a peripheral device slung off the backplane of a mainframe.
This is the long way of getting to the point of this week’s column. This week, I was introduced to a new technology from Zetera, a company you may never have heard of, but one that I wager could become a household name if they play their cards right. From the background provided by their PR representative, I am catching on to the Irvine, CA-based company a bit late. IDC already did a paper on them and other publications have given them several column inches. At that point, their technology was still on the drawing board. Now it’s real.
Zetera says that network storage should, well, use the network. You cable disk drives directly to the network using devices called tailgates. Firmware gets the drive an IP address. You then install device drivers on the servers and desktops that will have access to the drive nodes. Voila! You have a hugely scalable array, created by means of the IP network itself.
Bad news for the RAID guys: Zetera says they are the last remnants of a bygone era. Using IP multicasting and a bit of proprietary magic, the vendor can stripe data across disks, producing the same results as RAID 1, 3, 5 or 1/0.
Zetera handles storage I/O using UDP rather than TCP. TCP is favored by the Internet Engineering Task Force, ostensibly to guarantee interoperability between all products that adhere to IETF standards. That’s why TCP is used for iSCSI. The question that Zetera asks (as have many others before them) is why TCP, with all of its overhead, is required?
IETF might respond, “For security and reliability in hostile networking environments.” To Zetera’s way of thinking, this is all well and fine, but most storage will be deployed in non-hostile environments—inside data centers or in user departments on networks that are segregated from the general purpose LAN. In such environments, UDP is more than adequate.
A bit of history: back in the late 1990s, when iSCSI was being delayed by vendor obfuscation, a number of forward looking vendors, including Adaptec with its EtherStore product line, were preparing to implement UDP-based versions of network storage protocols. Cajoling by IBM and Cisco made them drop the idea and “fall into step with the mainstream” or TCP-based development effort, and TCP-based iSCSI took its dear, sweet time getting to market, giving FC vendors the early lead in the market.
Now that the big cats have lost some of their luster (Cisco trades for $18 per share, not $80-plus, and IBM recently reported an abysmally poor quarter), the idea of UDP-based network storage has resurfaced with Zetera leading the charge.
Will Zetera’s technology play well with those who like brand-name logos on their designer array cabinets? Probably not. As one reader pointed out to me today when I was talking about writing this column, “It will be difficult for them to break into the Enterprise market though, as the current vendors will FUD them to death.”
The same reader, however, pointed out two strengths of Zetera’s technology that might give them a place at the Enterprise table: 1) using the multicasting function of networks could enable the first true storage load balancer, which would simplify how capacity is managed today; and 2) wedding Zetera technology to extremely dense disk drives, such as Hitachi’s forthcoming 1 Terabyte perpendicular drive, may well send contemporary NAS and SAN solutions into an IT museum in short order.
Zetera doesn’t make storage devices. They aim to license their technology to others—a smart move. Initial targets for Zetera storage are the consumer and the small to medium business market where easily installed, readily managed (and cheap to buy and own) storage is much in demand. They have promised me a couple of copies of their wares for testing in my labs this summer. I’ll let you know what I find.
As always, your comments are welcome. firstname.lastname@example.org
Jon William Toigo is chairman of The Data Management Institute, the CEO of data management consulting and research firm Toigo Partners International, as well as a contributing editor to Enterprise Systems and its Storage Strategies columnist. Mr. Toigo is the author of 14 books, including Disaster Recovery Planning, 3rd Edition, and The Holy Grail of Network Storage Management, both from Prentice Hall.