iSCSI HBAs: Is this the end of the road?

In 2003, right after the iSCSI standard was ratified, the first iSCSI HBAs that hit the market were the Intel Pro/1000 T IP Storage Adapter and the Adaptec 7211C/F.

The basic premise of the iSCSI HBA has been to offload TCP/IP and iSCSI processing overhead to the card thereby freeing CPU cycles that could be used for greater processing. It was a good thought, however, in hindsight, it was based on the assumption that most servers were already running at high CPU utilization rates.

Five years and thousands of iSCSI implementations later, that simply has not the case. In fact the market started to validate this very early on by simply not adopting these HBAs at the rate their manufacturers had hoped. As a result of that, in July 2005, Intel issued a cease and desist order for further iSCSI HBA development and discontinued production due to "changes in market segment conditions". Adaptec followed suit shortly after that.

Around the same time, 2004, Qlogic announced their own iSCSI HBAs, the QLA4000 series. In fact, to date, Qlogic remains the largest iSCSI HBA provider in the market. Furthermore, variations of Qlogic's iSCSI HBA are also found on some iSCSI Storage Arrays serving as iSCSI HW Targets. Being the largest and most reputable iSCSI HBA supplier in the market and with the ability to boot servers of an IP storage device, Qlogic did OK but I don't believe that at any point they revenues exceeded their expectations. They did particularly well in environments, where iSCSI Booting was a requirement simply because there was not an alternative at the time.

However, over the past couple of years, with the introduction of SW solutions enabling iSCSI remote booting using regular NICs like winboot/i as well as Intel's and Broadcom's more recent approach to iSCSI booting using an embedded an iSCSI FW image in the NIC, the need for iSCSI HBAs has diminished at a rather rapid pace.

In fact, servers today pack more CPU power than ever before and can easily handle the TCP/IP overhead of an iSCSI SW solution with plenty of cycles to spare. That, and the fact that Microsoft has done an absolutely fabulous job with their iSCSI SW initiator and provided a special iSCSI initiator for iSCSI booting, has also played a substantial role in most deployments favoring iSCSI SW initiators over iSCSI HBAs.

For the most part, early ESX server deployments also used iSCSI HBAs. However, the past 12-14 months we've seen a significant drop in iSCSI HBA deployments for ESX server implementations in favor for the SW initiator. VMware's enhancements of their iSCSI SW stack has significantly improved performance, due to, among other things, increases in Virtual Adapter and LUN Queue Depth maximums. To date we and our partners are seeing the vast majority of iSCSI ESX server deployments occur using the ESX iSCSI SW initiator.

So is the end for iSCSI HBAs near? From an iSCSI initiator perspective, the odds are certainly stacked heavily against them and the introduction of FCoE will certainly seal the deal in the next few years.