Would you talk about future IP developments and any advice you could offer IT organizations still trying to sort out the various storage options in this field?
Here's the paradox. Let's look at this a couple of ways. In our research, we've found that about 89% of all servers are currently backed up over the network. A huge percentage of those are just on Ethernet subnets.
So, when we move to IP storage that has the intelligence -- to not just take backups in the sense of moving SCSI blocks --but to also take command and control (which is the IP Tape initiative), it's going to be an easy switch for these guys. All they really need to do is drop in a couple of NICs to give them this sort of added feature.
One of the pacing things in this industry is the NIC -- which is compatible storage and Ethernet NIC. There's a lot of work going on there. I think one of the biggest myths is that it's just an IP NIC. The reality is that moving storage is a lot more complicated than that. The networking companies that think they're getting into the IP storage NIC business are in for a surprise.
So, my prediction right now is going to be the Fibre Channel host bus adapter companies that are best positioned to pull this off in the early days, because they have that experience. Can you elaborate more on IP storage, and the road to its adoption in comparison to Fibre Channel SANs and other networked storage alternatives?
This is a paradox, because we are in the market-hype stage of this product development and like usual, we are trivializing how simple or hard this is. The point I'm getting to, is that [adoption] is going to be later rather than sooner.
I want to qualify this in two dimensions. We have switches being announced by a number of companies. But the paradox is, there are no bus adapters. So you can't really build a system. That's one side of it.
The second side of it is interoperability, infrastructure stuff...all that's not worked out. No, we don't have the standards done yet in IETF. Does that matter? Yes and no. If you're waiting for a proprietary system today, you can go buy a complete proprietary system today and guess what? It works. So, for example, the guys who really lead this market today are CNT because they're dealing with wide area network communications, and you create a gateway and a gateway and you're done. CNT's got their hands around that just fine. So, that's working in their environment. So, that's a simple case where you can do it, and you don't really care about the subtlety of the protocol or compatibility because it's not an interchange environment anyway.
The second piece I would make is that I believe IP storage will have a more rapid than normal adoption rate because it's coming on the heels of the effort of the SAN community. You mentioned a historical report and analysis you'd recently done in this area for Strategic Research Corp. Could you share some more of your findings with us?
Sure. I built an historical analysis of the penetration rate of various bus and network technologies as a relative percentage of new server shipments. So, we caught a picture of how many ports, how many buses, etc., relative to new server shipments. When you go back over time, we found that it typically takes 5 years for a new bus to go from introduction to about 20% market adoption. The same thing happened to Fibre Channel. The same thing happened to PCI.
Now let's think about the IP storage adoption rate. It's realistic to think that it would happen faster because it's going to be leveraged off the SAN and it's leveraged off the fact that 89% of the servers are already attached to a network for backup. These are easy transitions in both cases. Backup's the killer early adopter of IP storage, because of this 89 %.
So, you plot this out. I'm forecasting that Fibre Channel plateaus at about 40% market penetration, which, by the way is great. We never expected 100%. These poor folks that are running around saying, 'The sky is falling,' never looked at a history chart. So, 40% is great. It's really great. I mean, that's two times where we are today. So the Brocades of the world have got a great run of Fibre Channel coming ahead. You don't envision a one-size-fits-all storage solution in the offing?
There never will be. If you look at the new thinking of the switch technologies (Nishan's done a great job here), it's a common port and that port can handle a CAT-5 connector or whatever. It doesn't care what's on the wire. It could be IP packets. It could be Fibre Channel. It doesn't care. And it'll do internal translation switching and translations required.
Now move this up a layer to the application space. We already have products on the market called information servers. An information server is both a critical translator, a data type translator, an XML translator -- data in/data out -- and we can share information. The same analogy flows down to the protocol layer. We have a number of devices coming to market that are bridges -- protocol bridges. We have new type of edge devices to enable this. So, we're going to plug storage direct to Infiniband. It'll be plugged to a SAN. It'll be plugged to a MAN. And you'll just transmit whatever's appropriate. What do you make of the people who proclaim Fibre Channel is on its way out of popularity?
These naysayers just don't understand history so that they can't see what's going on in a bigger context. My prediction in time is that we have a common physical network -- optical channel -- and that common physical network connects buses between servers. It connects LANs. It connects storage. It connects MANs and WANs in many cases as well. It's a common physical layer -- optical channel -- is what we call it. It's not Fibre Channel. It's optical channel. And, we have logically different protocols. Which protocol we use is dependent upon the application we're trying to accomplish. End of story. It's that simple. Could you explain about the growing popularity of storage virtualization?
We're in this paradigm. In Infiniband, they call it peer-to-peer computing. In the SAN world, we call it serverless data movement. So, it's the notion that we're going to have external intelligence moving data, replicating data, protecting it, making it more accessible -- all those dimensions -- independent of the application servers. Therefore, we have to create this physical and logical separation. This explains the emergence of this virtualization era, so that we mask everything behind and can present a unified view of storage to the applications.
So, now you can start seeing the pictures emerging here and that leads you over to the AIN discussion, because we have to intelligently work with the applications, since the applications have no control over what's on the back-end. See, we separated it, but something new has to come along to control it.
So these are some fun new pictures for storage. IP storage will be adopted at a higher rate. I believe it will go much higher than 40%. But, I also believe Infiniband will take 5 years to get to 20%, just like PCI did -- even though other people might think otherwise.