The inflection point of 10 Gbit Ethernet is one thing that may speed adoption of iSCSI, but I would tell you that by looking at the performance needs of many applications the current 1 Gbit Ethernet standard is more than enough to handle the requirements of today and probably into tomorrow.
Think of a database or an e-mail application using the storage pipe, writing and reading to a storage pool. You would think of 8 Kbit or 4 Kbit transactions being written to and read from in a specific pattern, either random or sequential in a 1:1, 1:2 or 1:4 relationship. The storage pipe is just one important piece in understanding the total performance of a given protocol.
One must also look at the initiator or host, as well as the target, or storage pool itself. Is there enough horsepower on the host to push enough traffic down the storage pipe to throttle it and on the other end is there enough cache and disk spindles to handle the traffic coming in? The storage pipe is only a part of the overall equation of the total aggregate performance available. This is why many storage vendors provide tools to end users to help them understand the overall performance needs of a specific application so they can tune the right storage environment to their needs. In the end, the choice of a specific storage protocol can depend not only on performance, but past levels of comfort with a specific protocol, or even corporate standards for purchasing.
With regards to iSCSI specifically, you are correct that the protocol itself is being driven from the low to middle part of the market. At the high end, Fibre Channel is the predominant player. This is mainly due to cost and the perception the Fibre Channel is a better protocol. iSCSI has a lot going for it in addition to performance, including: ubiquity, interoperability, high availability and mirroring capabilities that are built into Ethernet, support by the leading OS platform and application vendors (including Microsoft, HP, IBM and others).
Also, 10 Gbit Ethernet is available today, but the costs are very high: $3,000 per NIC, and about the same for the port on a switch. However, most customers are using 10 Gbit for direct connections (no switch involved), so the cost to performance relationship is a bit off. But this will change rapidly, given the volume of 10 Gbit devices that are about to storm the market over the next 12-18 months.
In my humble opinion, 10 Gbit will help speed iSCSI along, but I believe that there isn't really a race here at all, rather Fibre Channel and iSCSI both have a place in today's IT realm, serving the needs of specific applications based on the requirements for the application. At this time of year, I would much rather think about harmony in the world rather than putting Fibre Channel and iSCSI at odds with one another. After all they are both talking SCSI, just over different transports.
Read Stephen Foskett's answer to this question.
This was first published in January 2005