Published: 08 Aug 2002
You've heard the term. You might even know what it means. iSCSI is one of the most hyped storage technologies in years, yet where is it in real life? I don't see it - and that bums me out - seeing as I was one of the early touters of the concept. I'd like to use this article to tell you what's happening, why it's relevant, why no one does it yet and what's going to have to happen for iSCSI to become relevant.
First, iSCSI is block data over TCP/IP (Ethernet, baby). Your computer talks SCSI to its disk drive, be it over a SCSI bus or a Fibre Channel (FC) bus (which is SCSI over FC). If you want to network your storage - and for the love of God you really ought to at least want to network your storage today - you either network for files via Ethernet (NAS) or for blocks via FC (SAN). iSCSI lets you network via Ethernet for blocks. But why should anyone care? Because 90% of the data storage in the world isn't FC enabled, but 100% of the world has Ethernet.
Here's what should happen
FC should continue to be at the heart of the data center storage network. It works, it's fast and for high-speed, low-latency, ordered transactions, it's as good as it gets. The problem with FC is that it's difficult, complex and expensive - and that restricts it to staying at the heart of the big data center. iSCSI (in theory) is cheap - you already have NIC cards and Ethernet - and pervasive. iSCSI should be used by:
- Those small- to medium-sized businesses and IT operations that have no FC, yet want to reap the rewards of centralized storage networking. That should be everyone. You will save a truckload of dough by centralizing all your storage in management costs alone - you'll be able to deal with at least ten times the capacity with the same administrator. I won't bore you with the other benefits - go look at any FC site and you'll find a hundred.
- Bigger businesses with FC who want to get even more centralized by using iSCSI to let their small servers around the enterprise access data off of the big FC SAN.
In either case, we're simply talking different scales of the same things. Centralized networked storage is the thing we all should be doing. iSCSI is an enabler for that to occur in a bigger way.
People who argue with iSCSI's merits are foolish. Cheaper technology that enables better infrastructures for the masses can't be bad. You can argue that it isn't as fast, as reliable, as pretty, etc., but you can't argue that it isn't good - in the right spots.
So why hasn't it taken off? Because of Cisco
Cisco in many ways has to be commended for the vision to see that iSCSI is a fantastic broad market technology. They spent a half a billion dollars buying iSCSI visionaries NuSpeed a year and a half ago. They recognized that storage was a fantastic way to fill up IP pipes, and in case you haven't noticed, Cisco makes a lot of money when people fill up IP pipes. They came out with a great initial product - the 5420 - enabling users to stick this box in front of their FC disk array and connect it seamlessly to servers running an iSCSI driver. Consumers would now be able to easily create much bigger SANs, further consolidating storage and easing management.
Nobody bought them. Actually, a few hundred did, but in Cisco terms, that's nobody. The pricing was way too high, but that's not the real reason. The real reason is arrogance and ignorance. Besides half of the 63 or some odd folks in Minnesota from NuSpeed, I haven't met anyone else at Cisco that speaks storage. They speak Italian, but they don't speak storage. Capisce? Prego.
To assume you will be successful in a business such as storage - just because you are dominant in a business such as networking - is silly. You need to put a little more than lip service into making a play in this space, like hiring a few serious storage execs, and maybe getting the sales force to push the stuff because it could be the most strategic move the company makes in the next five years. I think I could have sold a zillion dollars worth of this stuff by now, and I'm not that good.
Cisco just came out with the follow-up to the 5420 - the 5428 - which is even more brilliant than the original. They figured out that they wouldn't be able to destroy the entire FC industry with an Ethernet-only solution, so they incorporated a FC switch into the product. They dropped the price to a market leading level. In short, why would I buy a generic FC switch if I could buy the same switch functionality with 2Gb Ethernet iSCSI ports as well - for less money? I know I could sell that.
The problem again is what seems to be the "if I build it, they will come" ego that just ain't cutting the mustard. This is a killer product. If Cisco would only put a little bit of effort, it would not only reap great rewards for the company, but be a boon to the users who are desperately trying to get to a storage network without killing themselves in the process.
You hear that iSCSI hasn't been adopted because the TCP/IP offload engines - putting the dreaded TPC stack in hardware to stop killing the server - haven't appeared on the market in force yet, but that's not really true. There is a huge percentage of the market that doesn't require that level of performance to reap the benefits of iSCSI. When it does take off, it will only help. I love the things StoneFly is doing - they created a cheap (under $10,000) iSCSI gateway (like the 5420) that has a full set of storage virtualization and applications on it that makes iSCSI SANs have the full rich feature sets of many big brother FC SANs. StoneFly gives someone the impetus to get off the fence and get into the game.
What's going to happen? Assuming Cisco remains aloof to the game, the only way iSCSI becomes widely adopted is for one of the 800-pound gorillas of the business to realize that this is a good thing. Let's just assume that EMC/HDS/IBM/Compaq, etc. wakes up and realizes that while they are great players at the core, 80% of the data in the enterprise isn't attached to them, but every server in the company is attached to Ethernet. Wouldn't you think they would be excited to push the customer to stop managing direct-attached disk on their second/third tier servers - of which they have a billion of - and have the storage moved, real time, via iSCSI onto the big, fast, reliable back-end FC SAN? That helps the customer, and really helps the disk vendor. It adds a ton more capacity to their boxes.
As soon as that message gets pushed out we'll start to see some adoption. Within months the world could look entirely different, and we'll be bringing the benefits of a block storage network to far more places in the enterprise. As soon as a big dog starts the play, every little dog will follow. Pretty soon it becomes an iSCSI free-for-all. And that, my friends, is what's supposed to happen.
Finally, how do I know the market will eventually take off? Because Sun doesn't think it will. I love those guys, but let's face it - when is the last time they were right leading on a storage technology? My pal Balint, the CTO of Sun Storage and one of the smartest guys I know, has been arguing with me about iSCSI since the concept came out. So far he's right. I could use a little help here, people.
And now for something completely different ...
I applaud the inroads mighty Microsoft has made with its Server Appliance Kit (SAK) NAS operating system. It owns about 25% of the market, and is rising swiftly. What I don't understand is while they talk storage, they haven't done anything other than the SAK, which really is hidden by the OEMs anyway. When will they actually make a play with real data management or storage management products? There are a ton of startups with great stuff, and even a bunch of undervalued public companies that they could scoop up for a song to launch themselves into the middle of this market. Why not buy Legato? How about BMC? You get the point.
How about this one - forget backup. That's right. To hell with it. There will be technologies soon enough that completely eliminate the need for traditional backup, and hence backup windows. Imagine if every write you did was automatically mirrored and time stamped so that you could role back to a consistent state at any point in time? These are things that I think about. Loser.