I'm now starting to meet more and more classical hosting providers who want to evolve their business model. They've done pretty well in providing dedicated resources in efficient data centers at an attractive price.
Now, they want to move from a fixed-resource, fixed-price model to a more lucrative variable-resource, variable-price model -- presumably, where the variability is in an economically positive direction!
And many of them are starting to realize that the scale-out architectures that worked so well in their last model probably won't work out so well in their next model ...
The Essence Of Scale Out
On one level, scale-out seems to be the quintessential "lather, rinse, repeat" -- stand up small chunks of dedicated (or modestly shared) infrastructure as (presumably modest) customers come in the front door.
When a rack fills, buy another rack. When you run out of rack space, buy another data center.
And so on.
Efficiencies come from using standardized commodity components, or -- at least -- standardized infrastructure for the customer's choice of server, operating system and storage. Nothing wrong with this model.
But it has its limits.
The Seductiveness Of Scale Up
By "scale up", I mean one liquid pool of resources that can flex based on aggregate demand. Workloads can spill over to additional resources if needed (and customer has paid!). Or, if the resources aren't being used by the customer, they're available for other useful purposes.
First, any hosting provider (or service provider) that uses a scale-up model (big pools of resources) will have a fundamental economic advantage in cost-to-serve for variable workloads. And, trust me, almost all workloads are variable.
Why? You can oversubscribe your resources, meaning a lower cost-to-serve than the guy who's using a pure scale-out model.
But there's more to it than that -- you can tell your customers that vast pools of resource are available -- if needed -- without having to allocate them up-front.
That makes your modest infrastructure service even more attractive to any enterprise IT buyer -- it's a bit of insurance that we all would like to have -- and would be willing to pay a premium for.
So, for a given unit of IT workload, I'd make the argument that -- using a scale-up model -- that unit of IT workload is (a) cheaper to serve, and (b) more valuable to the customer.
Nothing Good Is Easy, Nothing Easy Is Good
But getting to this scale-up model isn't easy.
First, there are entirely new technologies to consider around virtualization of server, network and storage at scale. Second, managing QoS and security at high levels of multi-tenancy is still an embryonic practice.
And, finally, there's infinite complexity around policies -- and pricing -- around fixed vs. variable resource allocations.
Some service providers are looking at this brave new world as a "problem" to be solved, and tend to shy away. Others see a rich opportunity to differentiate and monetize their offerings in new and interesting ways.
Being the IT technology type, I tend to throw in my lot with the second crew. Besides, how much competitive advantage can you gain from doing exactly what everyone else is doing?
Digging Deeper
When we look at the multitenacy issues created by scale-up models, it can seem daunting.
How do you isolate resource allocation so that one tenant doesn't impact another? How do you set hardwalls (or softwalls) that provide well-defined QoS for individual tenants and groups of applications?
How do you convince customers that your multitenant environment is more secure (and more compliant, and more auditable) that the environment they're using today? And -- most importantly -- how do you create "control planes" for tenants so that they can monitor the day-to-day health and consumption of resources?
Many of the technology answers are starting to fall in to place -- at least, from my perspective. There's a bit more work to do, but you can see it from here. We're not that far away from being able to do all those things, and more.
This Shifts The Focus
Collectively, these newer forms of multitenant capabilities -- when considered in their totality -- can be evaluated one of two ways: (a) how do they help me do a better job at what I'm doing today, or (b) what new services and capabilities can be offered to my customers based on what the technology can now do.
Obviously, I'm much more interested in the latter than the former ...
Comments