This week has been great for reading the EMC blog-o-sphere.
First we had claims that our products don't work, and then we had Chuck Hollis' claim that our core technology was flawed.
What Chuck did get was that a NetApp core technology is the automatic optimized placement of data on a set of storage spindles. This was pretty impressive, because most of our competitors flinging FUD at NetApp don't get this. I prefer to use the term storage virtualization layer, but we'll use his. So Bravo Chuck!
Unfortunately, Chuck got off track.
- He dismissed the value of automatic data placement by saying that it could never perform as well as static placement,
- He argued that only NetApp and a collection of startups were doing it, and therefore it was not an orthodox architecture
- Finally he argued that Flash made automatic data placement irrelevant.
All three of his arguments are flawed. So let me tackle them one at a time.
Virtualization Can Never Perform
At the core of this argument is the belief that software can never be as smart and as efficient as a human being that carefully plans everything out with complete information.
The problem with this line of reasoning is that it is based on a flawed premise. A human can plan everything out as long as the system exhibits no dynamic behavior. Once you introduce any kind of dynamic behavior (i.e. have you had a spike in user activity or request to re-arrange data due to a re-org recently?), a human either cannot adjust fast enough or cost-effectively enough within a business-critical timeframe. In all of these cases continuous optimization software is always better. Since you can’t predict the future every day, no complete plan can ever be created. Automatic data placement is how NetApp addresses the dynamic nature of data centers.
But what I don't understand is how Chuck, who is a fan of VMFS, a host storage virutalization layer, who argues vigorously that not using virtualization for applications makes no sense, believes that storage virtualization within the array is a bad thing.
Chuck is right, doing automatic data placement is hard, which is why it took NetApp nearly 15 years to evolve that technology to become such a significant player in the storage industry. It's also why most startups have been unable to successfully compete with NetApp. Doing automatic data placement with few if any consequences is hard.
Not orthodoxy
Read my other blog posting on group theory and what it means for you.
Flash
Now here's where Chuck lost me entirely. He made this, bizarre argument that Flash obviated the need for automatic data placement because storage would be segregated.
Automatic data placement is a powerful technology within a storage tier. The value it brings to Flash is different than what it brings to disk but there is real value.
Now here's what I think Chuck was really thinking:
Automatic data placement is only useful for getting more performance out of spindles. No more spindles, no more need for automatic data placement, therefore technology is not valuable. Therefore NetApp not valuable. Therefore EMC wins....
But the problem, Chuck, is that automatic data placement is the foundation of how we do snapshots, and end-to-end deduplication, and FlexClones, and thin provisioning. Those capabilities add value regardless of Flash or spinning media. And for many customers the ability to turn all those features on without compromise is why they buy our products, often replacing compromise-ridden solutions from EMC, Dell, HP, HDS and others.
So Flash will not cause us to go away.
I know if you keep trying Chuck, you'll finally get what we do...

I like that our customers get it and chuck doesn't.
I also like the fact that you keep chuck busy thinking and talking about this stuff whilst we are out there helping his customers switch to NetApp.
Big fan of your work. Keep fighting the good fight!
Posted by: Aaron | August 13, 2008 at 12:37 AM