The iSCSI deciding factors

For the last 10 months, I've worked as the IT Director for Elmira College, a small liberal-arts college located in Elmira, NY, in the midst of the Finger Lakes, and one of the areas that immediately caught my eye was the storage situation in our data center.

With 35 servers, we have no central storage and no good way to implement highly-available solutions like clusters of servers. Moreover, each time we buy a server, we need to project the possible data storage needs over the next few years. Sure, it's doable, but we end up with a ton of wasted space, little ability to granularly manage disk space, and in the rare event that we underestimate our storage needs, we need to scramble to correct the situation. That's all about to change.

Thinking through the iSCSI vs. fibre channel decision
By this point, many of you reading this article are probably thinking that it's time we invest in a Storage Area Network (SAN), and, you're right.

Recently, we completed the selection and purchase of a new SAN. However, instead of choosing a "traditional" fibre channel SAN complete with FC switches and host-bus adapters, I made a decision early in the process to focus on iSCSI-based SANs.

Here's a summary of my reasoning:

1. First, I knew I wouldn't be able to present enough justification to the President's Cabinet to get an expensive fibre channel solution in house. Even with ROI and savings estimates—both cash and labor savings—it would have been a no-win for me, and I knew that.
2. Second, all of my staff is well-versed in Ethernet and TCP/IP, but no one—myself included—has had exposure to fibre channel, making the training side of the equation much more difficult. We're a small school with a relatively small IT staff, so adding major new technology to the portfolio can sometimes be difficult.
3. Third, I needed the SAN in short order, so extended training was not really an option. We're migrating to Microsoft Exchange over the next two months, as well as moving a number of vendor-supported databases to Microsoft SQL Server, and I wanted all new projects on centralized storage in order to provide highly-available, clustered solutions (Exchange, for example) as well as to have the ability to take regular point-in-time "snapshots" of our production databases.

I can't begin to describe the amount of research that I did prior to taking the iSCSI plunge. While I love new technology and like to see it in action, for my production environment, I remain somewhat risk-averse, and iSCSI is a fairly new technology of which I was very skeptical.

The first hurdle I had to leap was the speed issue. FC runs at 2 Gbps whereas iSCSI runs on top of Gigabit Ethernet links. In theory, this limits transmission speeds to 125 MBps for iSCSI and 250 MBps for fibre channel. However, with iSCSI's support for multipath IO (MPIO), we can use multiple network adapters to access the storage array. And, the simple fact is this: we have a small environment. We don't need massive storage bandwidth.

With iSCSI, we also have the option to use—but don't have to use—iSCSI adapters that improve throughput. However, to start with, we can use simple, standard Gigabit network adapters and determine from there whether we need more. Next, we don't need special, really expensive switches. A standard Gigabit Ethernet switch is all that's required. All in all, with the understanding of its potential storage limitations, the massive cost savings from iSCSI is due to its use of standard, inexpensive Ethernet hardware. --by Scott Lowe, TechRepublic