In-Depth
A Dirty Little Secret: The Industry and iSCSI
Disinformation in the iSCSI vs. Fibre Channel debates is stronger than ever
Recently, a headline appeared in a trade press publication proclaiming in boldface that iSCSI, the IETF standard protocol for operating block storage as an application across a TCP/IP network, was “not ready for business primetime.” The iSCSI protocol, according to the author, holds much promise for building low-cost storage networks. He argued that it lacked the support of a robust set of software technologies that were now commonplace in Fiber Channel storage fabrics—most important of which, in his view, being LUN virtualization.
This thesis was echoed by the commentary of an “expert”—a spokesperson for the “testing laboratory” of a well-known storage industry research and analysis firm. That source said that iSCSI’s immaturity and relatively small installed base had failed to stimulate storage software vendors to develop the necessary software management tools to support “enterprise class” storage solutions based on the protocol.
I was flabbergasted to read such words in a well-known publication—it was such an obviousexample of deliberate disinformation. “Analyst commentary for hire” being an established fact in this industry, there was nothing new about an analyst towing a client-vendor’s party line in a quote for the media. However, this was more than just an opinion expressed in an otherwise balanced news story. The entire story read like a plant: propaganda presented as news and deliberately designed to confuse or misinform the reader.
While I am not an iSCSI evangelist—in fact, I find myself put off by the nonsensical “one-protocol-fits-all” marketing claims of some iSCSI vendors to the same extent as I disdain the hyperbole of the Fibre Channel crowd—I found myself vexed by the blatant falsehoods presented in the story. How could such errors of fact creep through the editorial process?
It wouldn’t be the first time this had happened—some of my more curmudgeonly friends argue that it is standard operating procedure. Everyone needs advertising dollars after all.
To share my disgruntlement, I forwarded a copy of the article to the folks at DataCore Software in Ft. Lauderdale, FL, whose products have provided solid virtualization and storage management functions in iSCSI storage implementations for several years now. For the record, we have tested their SANmelody iSCSI virtualization software and management tools in our labs and have continued using them today because, for our application set, SANmelody works just fine.
I waited for a response to my provocation, imagining that the folks at DataCore would have smoke coming out of their ears after reading the piece. To my surprise, their CEO, George Teixeira, responded with the e-mail equivalent of a shrug and a smile.
Said Teixeira, “Wow, [iSCSI Not Ready For Business Primetime]—catchy title. But I think it’s a rehash of the major storage vendor, data-center-oriented viewpoint [that we’ve seen for some time now.]”
According to him, the vendors of FC SANs and large FC arrays are “out there pushing iSCSI double talk.” They tout their iSCSI strategies, but they aren’t pushing the technology in reality. Instead, they are “talking to data center types and ‘un-selling’ them from iSCSI solutions in favor of more expensive FC SANs or arrays—telling them that iSCSI is just a way to do ‘low priority storage,’ ‘non-mission-critical stuff,’ ‘near-line,’ ‘lowest storage tier,’ etc. … all terms intended to minimize the capability.”
Teixeira refers to the "bait and switch" technique: the vendor sets up a meeting to talk to a prospective client who is interested in exploring an iSCSI storage solution. Once in the room, the meeting is refocused toward the sale of higher-priced FC wares. Teixeira sees it as an effort to continue to milk the lucrative FC cow for as long as possible. In any FC sale, the purchase order usually includes one or more specialized FC host bus adapters for each server (at roughly $750 per unit), plus one or more FC switches (at around $750 per switch port), arrays costing between $189 and $144 per GB, plus lots of cabling, training, installation support, and maintenance contracts.
By contrast, selling iSCSI usually nets vendors the sale of an array or two. Strictly speaking, all you need is a network interface card, TCP/IP switch, a free software-based initiator, and an iSCSI array—that is, a standard array featuring iSCSI connectivity: typically much less expensive than a FC array. There is a lot less margin for the vendor or reseller in the iSCSI solution, even when you tack on virtualization software from DataCore, TCP offload engine (TOE) equipped NICs if you have an overloaded server, and a few other bells and whistles.
Still, the FC vendors find an audience for their propaganda, according to Teixeira. Data center-oriented consumers, who have long used channel-oriented storage architectures such as bus-and-tag, ESCON, and FICON to connect storage devices to processors in the mainframe space, are often quick to buy into “anti-iSCSI” rhetoric because the concepts of storage across a network are less familiar to them.
Fortunately, for Teixeira, not all storage consumers have a channel bias. iSCSI seems to resonate, he says, with the “mindsets and viewpoints of many networking (Microsoft and Linux network) administrators and users who are trying to solve everyday problems.” Network-oriented IT professionals “get” the value proposition of iSCSI immediately, just as they got the idea of network-attached storage a decade or so ago.
Being in the trenches, Teixeira should know. His quiet company had its best quarter in years at the end of 2005, adding noteworthy firms (such as retailing giant IKEA) to its growing list of SANmelody customers.
His analysis probably also rings true to Network Appliance and EqualLogic, both of whom claim to be the “iSCSI SAN” leader, though NetApp usually gets the kudos in analyst reports. Ask their people and they will tell you Q4 2005 seemed to be a turning point, with a lot of iSCSI product being sold to a wide range of companies, from Global 2000 players to mom-and-pop shops. Any suggestion that this technology is not ready to support mission-critical applications probably strikes their product sales managers as silly.
Sales trends are high and to the right according to John Joseph, VP of marketing at EqualLogic, who says he added 250 new customers in Q4, increasing reported customers for the company’s iSCSI arrays to over 900. Joseph reports a 40 percent re-purchase rate: 40 percent of his customers are so satisfied with iSCSI (and his company’s wares, of course) that they come back to buy more storage every year. Of his installed base, 90 percent use no TOE cards, no specialty switches—just software initiators. With the preponderance of apps in use today, especially Windows apps, a NIC is all you need.
That hasn’t stopped the Fibre Channel crowd from continuing to grumble about iSCSI at industry trade shows and in press accounts. Only a few months ago, a speaker for a major Fibre Channel switch vendor told an audience that iSCSI was a “good little technology” for connecting “unimportant” servers, those not already attached to an FC fabric, to the enterprise-class SAN—a simple gateway for connecting backwater servers to the corporate information repository.
While the public debate over the relative merits of iSCSI and Fibre Channel has quieted in the last year or two, it doesn’t mean that the relevant questions—those that will help you make intelligent infrastructure design choices—have been answered. Disinformation appears to have replaced evangelism in the vacuum created by the departure of iSCSI champions from the stage. As the more vociferous iSCSI advocates of the late 1990s, including Cisco Systems, shifted their focus to pursuing the lucrative low-hanging fruit of Fibre Channel sales opportunities, it has been left to smaller players such as DataCore to carry the ball.
Network Appliance’s new iSCSI marketing campaign will help, though it might also contribute to the confusion because NetApp represents its solutions as a kind of Swiss Army Knife—sometimes NAS, sometimes Fibre Channel fabric gateway, sometimes iSCSI SAN (whatever that is). For the most part, the important aspects of the iSCSI versus Fibre Channel debate has moved over into the Wild West of sales and marketing brochure-speak, where the most important credo remains “never trust nobody.”
We need more user cases—more “how I did it” stories about iSCSI implementations. We need to share real world experience to get past vendor and analyst folderol. Please send me your iSCSI stories, warts and all, and I'll report them here. [email protected]