« Who Can You Trust in Cloud? | Main | Ranting Without Tiers »

February 25, 2010


Hello Mike

A nice try, I must say.
You state Netapp is not as tier dependent as its competitors. It's a surprising way to put it. Tiering is a feature, not a constraint, and it satisfies the demand for cheaper data storage as information rolls forward in its life cycle and changes its value for the organization.

So let's get concepts fairly clear. I understand you may want customers to consider tiering as a low-value proposition. This is obviously related to product strategy and (this time yes) some architectural constraints. But let me say tiering makes a lot of sense.

I absolutely agree with the fact that different vendors have different approaches to market demands. Thankfully.

Now let me do a similar exercise with my experience with DOT. While other vendors (this is true) basically point at faster devices to gain speed (which, assuming data design is correct, sounds fair) Ontap points at more equal-speed devices to gain speed through spindle proliferation.

Well, it a way to achieve some gains, it's true.

But Mike, there's a downside for that one too (and I think you know it, btw).
So when you choose a manufacturer, as important as features are the constraints.

It's just a matter of choosing which constraints you would like to be facing and which not.



The end goal for both FAST/tiering and intelligent caching is to reduce the total cost of storage without sacrificing performance.

Customers could easily do away with tiers by buying all SSD, but it would blow out their budget. They use tiers as a way to push the cost of storage down. They don't want tiers, they want cheap storage that performs.

We are not chasing tiers, we are chasing customer demand. A future SATA + Intelligent caching approach is very viable for the majority of workloads and will have a very attractive price tag.

I'm not saying it eliminates the need for FC, at least not immediately, or the need to put some applications on SSD entirely. But, looking at the big picture from a customer's perspective, SATA and intelligent caching will be a huge part of their business. I'm sure NetApp will still offer the flexibility for those apps that demand more, but that will be a much smaller part of their datacenter.

The difference between the NetApp and EMC approach is that we have already solved the SATA random write performance gap by coalescing writes into sequential streams. This means as SATA drives get faster sequentially, we will be positioned to ride that performance increase. EMC is hoping for bursty write I/O they can treat with large write caches. The problem is that data eventually has to be written to the drive. Random data written randomly is slow, except on NetApp.

In regards to your comment about the mysterious, unspeakable downsides, feel free to share your thoughts. Otherwise, you start to sound like Hollis and that will lose you credibility here. Just be prepared to defend your comments with actual data.

Mike, hello again.

I think we have a protocol problem, I mean, we have a link, we both recognise ecachother as nodes... but somehow we don't get to connect.

There's nothing misterious in the downsides I mentioned. Furthermore, I think they're publicly known, it's just that I wouldn't like to mix discussions. Remember Ockham's blade, allways the simplest answer... ;), but we can get to it if you like. Bigger raidsets mean also higher chances for raidset failure, and an intrinsic difficulty to predict and control performance as different types of I/O patterns from different applications populate the raidset.

About your (i believe unintentionally) slighltly offensive comment on Chuck Hollis, I wish I had his view of the game.

Now, Mike, there's a difference between focusing a problem in real life, and doing it in the life you'd love to live.

The goal for FAST is to reduce administration costs when moving data between tiers. The goal for moving data between tiers is to obtain a reasonable cost for a proper application performance.
Now, what is "proper application performance"?, whatever the customer believes is good for his business. That changes (in real life, of course) from customer to customer. If a customer wants more performance in the same app, he'll have to pay a higher price per storage MB -today-.
If you doubt what I say, just check your own price lists and you'll find it's obvious.

Are you suggesting future devices in Netapp platforms will have more capacity and get faster?... wow! wellcome to the tech market. It better is that way, I want you guys in the business for a long time.

Just a comment about your SATA strategy
. What is the reason (in your belief) that drives customers to buy SATA drives? Let me put it some other way. You think they buy SATA for performance or for capacity?.

Cheers, Mike

The comments to this entry are closed.