Hybrid clouds, cloud brokers and cloud portability will all blossom as cloud computing trends in 2013, as long as important vendors don't stand in the way of that progress.
That's according to John Considine
The cloud providers are restricting the degrees of freedom for deploying services. It's just a reality.
CTO, Terremark Worldwide Inc.
Hybrid cloud seems to be everyone's prediction for how enterprise cloud computing adoption will ultimately settle out. What's your take on that?
John Considine: I used to talk about hybrid cloud as from a customer's data center to the public cloud, and that was the extent of hybrid clouds. Now, we look at hybrid clouds as a combination of colocation services, managed services, cloud-to-cloud services, as well as cloud-to-data center services.
When we look at how people use our data center today, there are a lot of people that will bring their databases to physical hardware in our colocation facility and connect with cloud. It gives them performance, security and better support from the database manufacturers that are still a little shy of virtualization in some cases.
What we'll see toward the end of 2013 and early 2014 is cloud brokerage, the ability to manage workloads across multiple cloud sites and providers. There's growing interest in the market to do this, and technology is maturing.
Inhibitors in this space have been the monolithic drive of the vendors. Suspect number one is VMware, where they want to do a vertically integrated stack, and provide the ability to go on-premises to off-premises, and manage workloads that way. The biggest challenge is that they want to have it be a sophisticated VMware stack on-premises and a similar or identical stack in the cloud, and it's an interesting form of cloud vendor lock-in.
They've done some things to expand their story beyond pure VMware, getting Dynamic Ops in there and acquiring companies like Nicira. That'll have a positive effect generically in the market. You'll see other big players, the BMCs and CAs of the world, trying to jump into the slot of a cloud broker, and then I think you'll see the providers, like us, trying to make it easier for customers to use these offerings.
Experts have said cloud portability is still science fiction today -- do you see that changing in 2013?
Considine: The cloud providers are restricting the degrees of freedom for deploying services. It's just a reality -- they don't want to do Layer 2 networking, they've defined their primary storage as single-connect, and the things that you can do are somewhat limited. Moving a VM [virtual machine] in its raw form is still hard if you think about rolling VMware into Amazon. The examples that people show are relatively trivial.
So we're seeing things evolve in the industry like software-defined networking (SDN). The other requirement is that the cloud service has the basic capabilities needed to be able to run. Moving forward, some of the work we're doing is to expand the control the customer has over the environment so they can make workload portability a reality.
Will you partner with Nicira or OpenStack for SDN?
Considine: We're doing our own thing there. People think we're crazy, but we had to. What we've learned from building very large clouds is that looking at some of the solutions that have been built for the enterprise that are trying to encompass a whole bunch of different people's problems, leads to problems for the service provider at very large scale.
We've invested in our own orchestration software, and one of our realizations was that [by] going down the path of SDN with a partner, we ended up with an issue about systems of record. What's happening with SDN manufacturers is that they're putting all their effort into the control plane, clusters to manage the network topology and the updates to the infrastructure, and we have a system of record for orchestration of the cloud that places all the workloads very specifically. So we ended up with two systems, one that was just taking care of the network overlay and one taking care of everything else.
As we looked at this, it didn't make sense. Since we're putting all the workloads in our infrastructure explicitly, we already know where all the endpoints are, and for us to do our own controller around SDN was really a much more economical solution.