Published: 14 Jul 2003
The low-risk route is through appliances--if they don't work, you can yank 'em. Easy, right? Cisco will put intelligence in its switch. Or I'll just run my Veritas software on my new Brocade switch, right? Well, maybe someday, but I highly doubt a lot of you are going to run to this new paradigm soon.
First, let me define what I'm talking about. Today we run virtualization in the form of volume management in the host, (usually Veritas') and LUN masking/mapping in the arrays. We use volume management to carve up physical devices into logical ones. We mirror with it and often replicate with an array-based technology. These are essential functions and perfect examples of "services" that should run in the network--not on the end points.
Why? Common repetitive software functions running in the network vs. the host mean far less software. If you're a big shop today, you run different versions of Veritas Volume Manager on your Solaris or Windows boxes, HP/IBM/SGI etc. If those core functions ran in the fabric, you'd support only a handful of platforms--vs. hundreds or thousands of servers.
Then, there's array-based intelligence: Move it into the network and you can eliminate the hardware tie-down. You can replicate anything to anything if the function is software in the network, not microcode on the box. In order to have true smarts, there has to be brains at every point in the wire, including in the network.
As much as vendors talk about putting the smarts in the switch, that ain't gonna happen in the real world for a long time (like maybe two whole years). For network intelligence to be significant, users would have to throw out the stuff they have (which works) and add new stuff that may not work. In the words of Dana Carvey doing George Bush Sr: "Not gonna happen." Human nature says, "Great idea, now tell me how to do it without risking life, liberty and the pursuit of a new career."
Instead, a better way for users to get comfy with this new intelligence manifestation is to incrementally add some smarts in a very benign way. "Network storage services" (NSS) platforms is the term we've coined to describe this. NSS platforms provide an off-switch way to add intelligence to existing networks, and easily remove it.
Some major players don't understand users talk big and walk slow. EMC thinks Cisco is "The One" and wants to port its cheap volume manager (I love this play) to Cisco switches. Are you ready to run all your VM functions on your Cisco switches yet? Thought so.
EMC should port to Troika or Candera's NSS platforms, and grab the entire fabric intelligence footprint. When the big guys are ready, and the market starts believing them, the market will want to run the software they already run!
Veritas is in the same boat, chasing the dinosaurs. Little guys like Incipient are more aggressive in supporting lesser platforms and will try to sell up when the market is ready. If I were EMC, I'd bury one of those platforms in my box and load up the VM, and tell the customer, "We're the first to deliver real network intelligence."
IBM is on board; its SAN Volume Controller is really an NSS platform with software. I'll argue HP's CASA replication engine is the same. It must have listened to users and figured out that baby steps are preferable to wholesale changes. Intelligence belongs in the network and on the switch. We just need interim steps to get there.