Whither iSCSI?

A May 4, 2001 eWeek article states that "iSCSI continues to replace Fibre Channel as the premier way to transport stored data." Then, seven months later, another eWeek article predicts that "new developments in storage over IP are signaling the end of Fibre Channel's reign as the high-speed storage interconnect."

And just to make sure I'm not picking on one particular publication, an October 29, 2001 a VARBusiness subhead predicts "Once the standards are set, IP storage is poised to come on with blazing speed." This was all printed while the iSCSI standards were (and still are) in development.

Now, let's fast forward to 2002. IBM recently--and quietly--stopped actively selling its pre-standard iSCSI-based 200i storage array. And Mountain View, Calif.-based 3Ware officially discontinued its PalaceAid iSCSI device. 3Ware executives were quite honest about the reason. Among other challenges, "it requires a lot of handholding and the performance was not there," says Mike Wentz, a 3Ware spokesperson.

This certainly does not signal the death of iSCSI. In fact, the iSCSI standards are slated for ratification by the IETF for sometime this year. And there are still vendors out there dedicated to making storage over IP work, especially on the host and networking side of the equation.

For instance, Nishan Systems, this week, announced that Carlson Shared Services, a solutions and services, company, deployed the first data center IP SAN using Nishan's IP storage switches--which function as edge switches for Fibre Channel storage or hosts. And companies like Adaptec and Alacritech have developed network accelerator cards that put the TCP/IP processing into hardware for reliable data transfers across the network.

The TCP Offload Engine (TOE) is considered a critical part to making this technology a viable competitor to Fibre Channel. Analysts say that without TOE, there is no iSCSI technology. Problem is , while there are companies working on improving storage over IP technology on the networking and host side, none out there are working on the storage (also known as the target) side of the system. 3Ware and IBM were the only two companies doing it.

What makes the fact that IBM discontinued its iSCSI box interesting is that Big Blue was one of the primary instigators behind the iSCSI specification. At least, that was the case more than two years ago when I wrote my first IP storage article.

Now, IBM's public response included this statement, according to IBM spokeswoman Kim T. Nguyen: "We believe this interface is more appropriate as a network-level interface vs. one to attach devices such as HDDs within a system. Feedback from our customers supports this direction, and it appears to be consistent with that of other HDD suppliers in the enterprise/server segment... But for now, the primary interfaces in server HDD are expected to be SCSI and FCAL, with the potential emergence of SAS in late 2003 to 2004 time frame."

It would be nice if IP storage would come through. Time and again, customers have told me that Fibre Channel is complicated and expensive. The well containing IT workers who specialize in Fibre Channel is less than half full. On the other hand, there is an abundance of IT employees who understand IP, thanks to Cisco Systems, which made networking a cool technology.

Now, the word out there is that IP storage is suitable in the low-end part of the market because some speculate that smaller business are more tolerant of low performance for storage traffic. And, as many of my Compaq and EMC sources predicted more than two years ago, IP storage will exist outside the data center--as a way to connect two SAN islands or to connect servers to switches with some kind of gateway used to convert IP into Fibre Channel for the disk array. Within the data center, Fibre Channel is still king.

I guess that is what we get when we try to make a new technology revolutionary when it is merely evolutionary.