Is iSCSI ready for prime time?

It's been three years coming, but there are signs that iSCSI may be ready for end-user adoption. Although the original developers of the spec—Cisco and IBM—have for the most part failed in spurring user interest in the protocol, support from two other heavyweights—Microsoft and EMC—may tip the scales.

Click here to enlarge image

Free iSCSI software drivers started rolling out of Redmond earlier this summer (see the cover story in this issue), and if "free" doesn't convince end users, then nothing will. (For an analyst's take on Microsoft's role in iSCSI adoption, see John Webster's "Which vendor will save iSCSI?", p. 46.) Users are expected to the use the free drivers mostly in low-end, relatively low-performance, low-cost (versus Fibre Channel) SANs.

Higher up the food chain, EMC next month will ship a multi-protocol board for its DMX disk arrays that supports iSCSI (in addition to FICON and Gigabit Ethernet). Can native support for iSCSI on EMC's midrange Clariion arrays be far behind?

Now that EMC's on the iSCSI bandwagon (as is Network Appliance), it won't be long before the other big RAID vendors follow suit. Expect iSCSI arrays from vendors such as Dell, Hewlett-Packard, IBM, and Sun in the fourth quarter or early next year. In fact, most of them could probably ship iSCSI arrays today, but why forego those Fibre Channel margins until you have to?

And within that same time frame, iSCSI host bus adapters (HBAs) from market leaders Emulex and QLogic will have reached end users (the HBAs are currently being evaluated in the channel), joining iSCSI adapter vendors such as Adaptec, Alacritech, and Intel.

If that isn't enough, Microsoft claims that more than 40 vendors are developing hardware and software products that will work with its iSCSI drivers.

If you're considering iSCSI, one of the first decisions you'll have to make is whether to take the lowest-cost route with just iSCSI software drivers and standard Ethernet NICs, or opt for more-expensive cards that offload TCP/IP and iSCSI protocol processing from host CPUs. For more information on this tradeoff, see "iSCSI software initiators vs. iSCSI HBAs," by Saqib Jang, p. 35. (Next month, Jang will examine the issue of using TCP/IP offload engines, or TOEs, to improve backup performance.)

4Gbps vs. 10Gbps

Meanwhile, Fibre Channel keeps chugging along, although it's unclear whether the next speed target will be 4Gbps or 10Gbps at the SAN level (see "FCIA approves 4Gbps SAN fabrics—Will switch vendors and OEMs follow?", July 2003, p. 12).

In a recent InfoStor QuickVote poll, 525 respondents voted for 4Gbps while 226 cast their vote for skipping 4Gbps and going directly to 10Gbps (the next speed for Ethernet). However, we didn't limit the poll to end users—vendors and integrators also voted, and we suspect that some of the more fervent 4Gbps supporters may have stuffed the ballot box. (Voting in QuickVotes is anonymous, so we weren't able to confirm that suspicion.)

If you're an end user and you have an opinion on this issue, drop me an e-mail (daves@pennwell.com) telling us why you think that 4Gbps or 10Gbps should be the next speed for Fibre Channel SANs.

Dave Simpson,

This article was originally published on August 01, 2003