The "Year of iSCSI"

I just read a blog by Chuck Hollis, from EMC, musing about why iSCSI hasn't been that successful. He says we keep waiting for the "Year of iSCSI", but it never comes.

I respectfully beg to differ!

For years, iSCSI really did stand still. The spec was coming, almost finished, almost-almost finished, shipping any year now... I do admit that's how it felt at first. But when I look at the iSCSI numbers for the past few years, what I see is an explosion of success.

Here's a little thought experiment. Let's compare the first four years of iSCSI revenue with NetApp's first four years of revenue. (I think that most people would acknowledge NetApp as a high-growth successful startup. :-)

	Year	iSCSI	NetApp
	1	$6.7m	$14m
	2	$80.9m	$43m      (The year NetApp went public!)
	3	$228.3m	$93m
	4	$457.5m	$157m

[Note: The iSCSI numbers are for the four-quarter periods ending in Q3, to align with the most recent IDC data. The NetApp revenue numbers are my fuzzy recollection and could be off by +/-$5m in the last two years.]

Just to put things in perspective, iSCSI is now ten times as big as NetApp was when we went public. I can tell you that it sure felt like the "Year of NetApp" when we reached that milestone! iSCSI's growth rate is way outstripping NAS in the early days.

I think that Chuck is thinking about it wrong because he is thinking from the perspective of a big VP at a multi-billion dollar company. Of course, I'm also a big VP at a multi-billion dollar company, but I remember NetApp's early days well enough to know how hard it is to create the first $100m in a new market.

My theory is that $100m qualifies any new technology for "the year of". (Back where I come from, a hundred million is a lot of money.) This means that "The Year of iSCSI" was 2004, with $109.6 million in revenue. The most telling statistic is that iSCSI has doubled in the last 12 months and shows no sign of slowing. The quarter-over-quarter growth rate was 15.6% three quarters ago, 19.1% two quarters ago, and 18.5% last quarter. Let it double two or three more times and iSCSI will definitely be a force to be reckoned with.

The other place I disagree with Chuck is his expectation that, "You'd think it'd be pervasive in large enterprise shops these days." A year ago I argued that, iSCSI Sucks, but that's missing the point. It's cheap and it's easy. My point was that you should not expect iSCSI to compete against Fibre Channel in enterprise data centers. Instead you should expect it to compete at the low end, helping small windows servers enter the modern world of networked storage. Over time, iSCSI might become disruptive and compete against high-end Fibre Channel, but that's not where you would expect the first billion in revenue.

In closing, however, I have to admit that I also agreed with many of Chuck's comments. In particular:

"Customers could save big money by avoiding the purchase of fibre-channel HBAs."

"They could use lower-cost ethernet switches rather than their more pricey FC counterparts."

"As we matched up iSCSI performance against our rather extensive knowledge of real-world workloads, it became pretty clear that a significant majority of applications could run comfortably on iSCSI with no negative performance impact whatsoever."

"First, I think you'll see more of the same: iSCSI in smaller, greenfield SAN builds where FC isn't entrenched yet. But from small acorns mighty oak trees grow."

In the end, Chuck and I may agree more about iSCSI's future than about its past. I'm arguing that the "Year of iSCSI" is well behind us, back in 2004, but as for the future of iSCSI, Chuck may be more right than wrong.

[The iSCSI numbers are from IDC's "Worldwide Quarterly Disk Storage Systems Tracker, Q3 2006 Release", November 2006.]

Comments

[Oscar Olsson]

As an employee of an outsourcing/hosting corporation, I was surprised that you were still skeptic to iSCSI deployments in enterprise environments! As you are apparently already aware of, the system owners always overestimate the I/O, bandwidth and CPU/memory requirements of their shiny new über-applications. In our current FC SAN, the average I/O load per LUN is like 0,1-0,5MB/s, and the peaks mostly happen during backups.

So, our conclusion is that iSCSI will cope with the I/O load in future deployments without problems, especially with the deployment of 10G ethernet, which we in turn plan to deploy during the next three months, so no issue there.

In the past, we have often thought about the huge gap in cost between DAS and SAN disk per GB, which made both us and our customers reluctant to invest in SAN storage. In my view, its pretty stupid to have several infrastructures to move data from A to B, regardless if its files, storage commands, or voice/media. Its costly and not very efficient. Why not invest the money that you previously put in building a POTS network, and a SAN network, into your IP network and deliver those services over the IP network instead?

AFAIK, there are few obstacles left to make this a reality with current technology. The only concern we still have left about iSCSI is LUN security. It would be bad for us and the customers, if they could mount eachothers LUNs. Some appliances still don't support LUN masking based on IP and/or authentication. In general, it puts higher demands on security when a device is exposed over a more far-reaching IP network, rather than a closed flat network such as a SAN. That ofcourse also includes scenarios such as buffer overflow attacks, DoS etc, even though the iSCSI IP is firewalled from the Internet.

But other than that, why not be more positive about iSCSI as a real competitor to FC SANs in 2007?

[Tony Asaro]

Hi Dave,

I think you are spot on except for one point. But I am not going to tell you what it is. You will have to read my blog instead ;-)

http://esgblogs.typepad.com/stor_wars/2007/01/i_just_read_chu.html

[Curt Beckmann]

Yes, iSCSI is a true space, and no longer stands for "I still can't sell it". But iSCSI is a long way from meeting the early unrealistic promises. The burden that iSCSI bears is derived from its outrageous early hype. It was going to be all things to all people. iSCSI in the home, kitchen, bathroom, HPC, etc. Nothing can live up to that level of hype, of course. On the other hand, iSCSI's beginnings happened during a recent "Age of Hype" knowns as the dot com boom. If you didn't hype the heck out of something you couldn't get any attention. Several other nascent technologies are similarly working to shed the excess hype of their early days.

[Rob Baker]

Dave, these numbers truly speak for themselves. We made the commitment to iSCSI for our tactical MS Exchange farms on NetApp in early 2004 and have really enjoyed the results. We just didn't want the cost or complexity of FC on Clarion in the desert and needed to push the innovation envelope. Our tests showed that iSCSI performance on NetApp was almost as good as FC on Clarion so we took it to production. Thanks for the innovative leadership in this important network protocol standard ... NetApp producing iSCSI-based Storage Area Networks has become a less costly but worthy alternative to creating Fibre Channel-based SANs.
Semper Fi and keep up the great work.

[SteveT]

I grew up where ESCON/FICON/FCP were barely capable of handling my IO demands. Today, iSCSI via NetApp is working in so many places I wouldn't have dreamed possible in 2004. From MS Exchange Farms to VMWare and MS SQL Clusters, iSCSI is performing well enough to satisfy. 10G ethernet, iSCSI HBAs, IP guys finally figuring out FULL DUPLEX/1000 Hard Coding. . . reality. Let's see 2007 double that last quoted number of 457.5 million with 2008 pushing us into a 1 billion dollar iSCSI spend.

Dave! Glad you went live with comments! Thanks for sharing your vision; keep up the great work delivering awesome products and innovation!