Internet Small Computer System Interface (ISCSI) is one of the hottest topics in storage networking. The first of three protocols designed to allow Storage Area Networks to rely on TCP/IP for data transmission is being ushered in on a wave of products, an almost-completed technical standard and a tidal wave of articles, discussion and just plain hype.
The idea of using well-understood protocols like TCP/IP and less expensive equipment to get the benefits of storage networking is attractive to enterprises of all sizes. Even after several years of wide adoption, the protocols underlying Fibre Channel SANs are still considered esoteric, while TCP/IP and Ethernet -- even Gigabit Ethernet -- are more likely to be seen as old friends.
However the real question most enterprises have about iSCSI today can be summed up in a jargon tagline the computer community lifted from an old commercial: "Is it soup yet?" In other words, can an enterprise install an iSCSI storage network today and expect it to work effectively in the real world?
The answer, unfortunately, is "it depends." Part of what it depends on is your definition of 'soup' For some people Fibre Channel SANs still aren't soup in spite of tens of thousands of installations and years of effort because there are still a few nagging problems. Definitions aside, there is the fact that iSCSI still has some issues to be resolved. Whether that keeps it from being a solution you should consider depends on how important
For starters, let's note that you can purchase a working iSCSI storage network today and mix components from different companies, although not necessarily every possible combination of components. (Which, to be honest, you still can't do with Fibre Channel SANs either.) Companies in the iSCSI industry have demonstrated working iSCSI SANs with components from different manufacturers at a number of almost feverishly publicized events at trade shows in the last year. What's more, iSCSI manufacturers have taken a lesson from the Fibre Channel SAN experience and worked hard to narrow the areas of incompatability between products right from the beginning. As a result, a growing list of products will work together effectively and interoperability is much less an issue than it was for Fibre Channel SANs at a comparable stage in their history.
The major consideration is that for all the manufacturers' support and all the products being released, iSCSI still isn't a formal standard. It is in the last stages of release by the IETF, which means things could still change, locking early adopters out of the iSCSI mainstream.
As a disadvantage, this is more theoretical than real. The iSCSI standard has been unusually well thrashed out and isn't likely to change much in the final phase of its adoption. What is covered in the standard being currently discussed is almost sure to be what is finally adopted.
A more serious problem is that there is still no standard for Windows 2000 support for iSCSI. That means that the current iSCSI adapters aren't certified by Microsoft, which is a barrier for many enterprises that require equipment be certified by Microsoft to work with Windows. It also means various manufacturers are having to develop their own, not necessarily compatible, interfaces for Windows and they may all become obsolete when Microsoft does release an 'official' version.
Another problem is that the first-generation iSCSI network interface cards (NICs) don't support common SAN features such as load balancing. This is because iSCSI NICs offload some or all of the job of protocol handling from the CPU. They do this because TCP/IP has an enormous amount of overhead compared to a protocol like Fibre Channel, and using it in a storage network tends to drag down even the most powerful processor. This is fine, except it means that the NIC's TCP/IP stack needs a special driver to communicate with the operating system's TCP/IP stack, which will lead to a plethora of different, not necessarily compatible, NIC drivers.
The best answer to the question "is it soup yet?" is that it is not nearly as soupy as it will be in another year. Storage administrators who decide to hold off on adopting iSCSI will get a better, more interoperable product than the early adopters who are rushing into iSCSI now. However, please note that none of these problems is insoluble, or will even take particularly long to fix. Intel and Adaptec, for instance, have already promised NICs by the end of the year. If you're aware of them and move cautiously, you can have a functional iSCSI network today.
An overview of IP storage generally, including iSCSI is available in a Dell Computer Web site.
The IETF memo on iSCSI is available from the IETF Web site.
Rick Cook has been writing about mass storage since the days when the term meant an 80K floppy disk. The computers he learned on used ferrite cores and magnetic drums. For the last twenty years he has been a freelance writer specializing in storage and other computer issues.
This was first published in August 2002