iSCSI struggling to live up to its promise

When it arrived, iSCSI (Internet SCSI) sounded not only exciting but irresistibly logical. Everything else has been hopping onto an ever-faster Ethernet/IP bus, so storage traffic seems to be a logical new passenger. Using IP to move block-level storage traffic via Gigabit Ethernet apparently gives CTOs everything they want, including increased competitiveness via new technology; continued leveraging of existing infrastructure; and a low, possibly nonexistent, learning curve for IT staff.

Despite these advantages, the 2003 InfoWorld Storage Survey revealed that only 13 percent of readers have managed any kind of iSCSI adoption. The answer to the mystery of iSCSI’s slow adoption lies both within the technology as well as the minds of potential adopters and their new environment.

“CIOs simply can’t afford to make potentially expensive mistakes in the current [corporate] climate,” says Jeff Allen, CIO of Glenwood Management, a property management firm managing a $5 billion Manhattan portfolio. Committing corporate dollars today is dangerous territory for most chief technologists. Continues Allen, “I’m not looking as much for a big technology touchdown as I am for a safe bet that still moves my company forward. iSCSI simply isn’t that safe a bet.” Looking closely at iSCSI reveals some truth to this assessment but also shows some imminent relief.

Where’s the standard?

If anything makes IT managers fidget, it’s getting involved with proprietary technology. And although iSCSI as a concept seems built on three established standards — SCSI, IP, and Gigabit Ethernet — the reality of melding these three technologies at present requires some proprietary pushing. That’s because an IETF-ratified iSCSI standard still hasn’t seen the light of day after more than a year since the technology’s introduction.

While this hasn’t dampened the enthusiasm of new and second-tier storage players, including Alacritech, Cenata Networks, and EqualLogic, it has seriously slowed product releases from the storage industry big boys, especially EMC, Hewlett-Packard, and IBM. Big Blue is even cited by some as one of iSCSI’s initial failures because it introduced its iSCSI-capable storage array TotalStorage IP Storage 200i, but then became the victim of industrywide rumors that the product had been canceled. IBM denies this rumor, and the product is still available on its Web site (see http://www.storage.ibm.com/snetwork/iSCSI/index.html), but reports say that customers’ ability to purchase a TotalStorage 200i remains hampered.

The good news is that the IETF’s IP Storage Working Group has promised an approved iSCSI specification by the end of March 2003. Lack of this specification has limited iSCSI products mainly to iSCSI software drivers and adapters. These products allow iSCSI implementation by sending iSCSI traffic to compatible target arrays (of which there are few) or, more likely, via compatible IP storage routers from vendors such as Cisco and Stonefly Networks. IP storage routers generally act as bridges between iSCSI, Gigabit Ethernet, and Fibre Channel.

Unfortunately, this missing specification has caused problems for most of these devices because the need for a proprietary technology boost forces some iSCSI adapters to assume the full function of an IP stack within their own silicon instead of leaving it with the operating system, where it usually runs. Called TOEs (TCP/IP Offload Engines), this style of adapter has run into routing problems, especially when determining efficient routing paths or handling load-balancing tasks in concert with Gigabit Ethernet adapters elsewhere on the network. Experts also say TOEs will suffer from compatibility problems between disparate iSCSI and Gigabit Ethernet product lines, although no true iSCSI intervendor compatibility testing has been done to date.

Thus, even though linking iSCSI traffic via a cobbled solution of adapters and routers is certainly possible, it’s not the vision that made the technology so popular at its introduction. Routing IP storage traffic over multiple underlying protocols is exactly what iSCSI was supposed to avoid. The attraction lies in the theory that, by using iSCSI, existing SCSI storage management software could extend over existing Gigabit infrastructure or even the Internet and manage existing storage devices. Leveraging all of this existing technology provides not only a low-cost implementation but an easy platform for future innovation in areas such as management software and SAN security protocols.

Using many new and proprietary technologies to manage a bridging process between iSCSI and several other storage protocols simply brings the SAN administrator back to square one. It’s also in direct contrast to IT leaders’ goals for storage technology over the next 12 months. When we asked survey respondents this question, an overwhelming 68 percent cited the growing importance of secure backups as a primary storage technology motivator. Adding additional layers of technology, complexity, and failure possibilities, as do many current iSCSI implementations, is not conducive to a secure backup goal.

A quiet front row

Aside from a working specification, iSCSI success requires native iSCSI storage products from recognized players, especially EMC, HP, Hitachi, and IBM.

While IBM insists that its TotalStorage 200i is still available, other top-tier players have simply maintained silence about iSCSI product releases, again relegating CTOs to lesser-known brand names; not something they shy away from for established technologies but certainly a negative factor when implementing something new and unproven. This was borne out in the Storage Survey, in which between 50 percent and 71 percent of respondents say they would consider a top-tier storage solution purchase within the next 12 months, while only between 4 percent and 12 percent say they would consider a solution from a second-tier player.

Fortunately, the announcement of a soon-to-be-approved specification has breathed life into the product lines of some of the big players. Hitachi, for example, has announced native iSCSI support in its Freedom Lightning 9900 array, EMC may be close to announcing similar support in its midrange Clariion line, and HP has already announced switch and router support for its StorageWorks line in the first half of 2003.

Alphabet soup

Finally, as with many new technologies, innovation around IP storage has moved so rapidly that it may actually have had a dampening effect on adoption. Although routing storage traffic over existing infrastructure is iSCSI’s raison d’être, that doesn’t mean it’s the only game in town — and certainly not in the traditional SAN market.

The debate over whether iSCSI and FCIP (Fibre Channel over IP) can live in harmony still rages on, although a close look at both technologies seems to indicate a certain complementary dichotomy. Given its inherent parameters, iSCSI seems well-suited to low-cost local storage connections. A key reason is that iSCSI generally requires end-to-end TCP/IP support. FCIP, on the other hand, is more expensive in theory but currently may be cheaper to deploy than iSCSI if it’s used as a long-link interconnect between geographically dispersed Fibre Channel SANs. In this scenario, FCIP easily integrates with existing Fibre Channel installations, making it quicker and easier to deploy than iSCSI.

A key factor here is storage management software. Even companies that are using large external storage arrays may not be employing SAN management software. According to the survey, only 33 percent of respondents are using SAN management software today, while 21 percent are still researching such software and 19 percent aren’t considering it at all. Given iSCSI’s present need for an additional software layer, even if only in a bridging capacity, it’s easy to see why IT executives might be shying away from such solutions right now.

Another player often mentioned in conjunction with iSCSI is InfiniBand. Existing product announcements for InfiniBand, however, show the technology being adopted mostly by large server vendors for use as a server cluster connection. It’s also been tapped as the eventual successor to PCI (Peripheral Component Interconnect) for local disk connectivity. In that role, it will need to cooperate with iSCSI but certainly won’t supplant it.

So,  iSCSI’s slow adoption really has little to do with competing standards. A viable growth path exists, but the lack of an approved standard has crippled this technology’s maturity over the last year. It hasn’t been able to prove itself, which means customers still face a certain amount of additional implementation risk — and that’s too much risk in the present corporate climate. To become successful, iSCSI needs an approved standard, native brand name support, and some loudly trumpeted customer success stories.