On your mark, get set, go iSCSI!

A number of vendors have been pitching iSCSI storage solutions to me. I thought the iSCSI specifications were still in flux. Should I factor iSCSI into my purchasing decisions?

If you are married to the belief that an industry standard isn't "real" until all products comply with it, then iSCSI is not for you—not now anyway. As of press time, the final draft of the iSCSI specification had not yet been ratified.

What does the vendor community have to say about implementing iSCSI products? Well, they are split into two camps. While some storage vendors believe the current draft of the iSCSI specification is close enough to final to begin manufacturing and selling iSCSI products, others are waiting for a final spec and for market demand to build before they ramp up production.

Where do I stand? I fall into the first camp and believe the specs are close enough to completion to begin planning iSCSI deployments. In fact, I would go so far as to put an iSCSI solution into production today, provided that component interoperability had been demonstrated.

After all, the entire history of SCSI and Fibre Channel is marked by products that come out ahead of specifications or by vendors that innovate outside the parameters of industry standards. The conventional wisdom has always been to assume that no two components work together. Even with relatively mature Fibre Channel standards, we rely on vendors or third parties to certify interoperability among devices, often down to the firmware version numbers.

In other words, standards are an important part of the picture, but, for any emerging technology, they are only part of the picture.

The good news about iSCSI is that it is based on TCP/IP and Ethernet—two very mature standards with near-universal interoperability. That means that network interface cards (NICs) and switches are not likely to be points of incompatibility.

The other good news is that there are relatively few iSCSI initiators (i.e., software drivers, TCP offload engines, and storage NICs) on the market, so target device (e.g., RAID, storage virtualizers, and tape) manufacturers can fairly easily assure end users that their products will work with a good subset of initiators.

My team has been tinkering with iSCSI technology for the last few months. In that period of time, we have had several firmware flashes and driver updates on the host side and one or two updates on the target side. But the rate at which new updates have been coming out has slowed, indicating that things are getting increasingly more stable. Almost all of our testing has been on Windows 2000 hosts, and we have been very happy with the results.

In fact, we are seeing performance that is comparable to Fast-Wide SCSI and Ultra Wide SCSI, which is perfectly suitable for just about all Windows and most low-end and midrange Unix servers.

We're seeing speeds on the order of 6MBps to 8MBps—with 10/100 Ethernet. This is suitable enough to drive a DLT8000 or AIT-2 tape drive, and it does not mandate the use of a TOE or SNIC on the host computer. Thus, sites equipped with a simple Ethernet server backbone can realize the benefits of storage area network (SAN) backup simply by loading a software driver on the hosts and connecting a tape drive or two to the Ethernet backbone via an iSCSI SAN router or bridge.

And we've found that the time, energy, and expertise required to build a small iSCSI SAN is comparable to that of configuring host-based RAID arrays for each server. This analogy is based on the assumption that you have a dedicated Gigabit Ethernet switch, or that someone else has taken care of configuring a VLAN on a Gigabit Ethernet switch. In other words, if the switches are ready to go and if IP addresses have been assigned, configuring a small iSCSI SAN is quick and easy and should be within the skill set of any technician familiar with basic networking and storage.

The one shortcoming is that we haven't been able to boot from the iSCSI storage device. Once we're able to do this, I believe the out-of-pocket costs of the iSCSI SAN should be comparable to the cost of multiple direct-attached storage systems. In other words, end users will be able to buy centralized storage systems without having to subscribe to complex ROIs.

Also it's important to note that the certification testing required for iSCSI tape solutions has not yet been formally qualified, which means you can't install iSCSI for the purpose of tape backup and expect to get tech support or endorsement from tape hardware or backup software vendors. The good news is that iSCSI tape solutions appear to work seamlessly (at least in our tests) with existing tape hardware and software. So, all we really need to move forward with iSCSI tape is an endorsement from the backup system vendors.

All and all, I am very excited by iSCSI. I believe iSCSI will drive mass adoption of SAN technology. The biggest pain points—storage provisioning and tape backup—are addressed very inexpensively using iSCSI, and more-advanced features such as mirroring, replication, and snapshots do not carry the same premium that they do with Fibre Channel equipment.

Bottom line: IT managers should begin planning for iSCSI, and those with the greatest pain should consider diving in today.

Jacob Farmer is the CTO of Cambridge Computer. He can be reached at jacobf@cambridgecomputer.com.

This article was originally published on January 01, 2003