Interop NY: The Big Migration: Moving the Data Center to On-Demand

Posted September 18th, 2008 by Joe Pendry

StackSafe CTO Drew Gross participated on a panel at Interop New York this year in the cloud computing track.

Moderator: Alistair Croll, BitCurrent
Andrew Gross, StackSafe
Bert Armijo, 3Tera
Michael Crandell, RightScale
Stuart Carlton, Elastra

IT’s About Uptime pulled a summary together based on the discussin during the panel:

One of the big challenges with cloud computing is that many IT people are told to embrace it, but do not really know how to do it. There are more challenges for the enterprise than for Web 2.0 companies.

StackSafe is providing a unique view of the cloud, as “testing is a gateway drug for cloud computing”. Companies are scaling from 1,000 servers from one day to the next. Many CIOs see testing and training as two of the areas where a company and enterprise can embrace cloud computing first.

How will enterprises eventually move towards on-demand computing? Will it be through SaaS or clouds? There’s a lot of customization that SaaS has not been able to do yet. On-demand allows for more scalability. A lot of people are doing migration on old equipment and data centers so they move to the cloud. The cloud has great value for enterprises in aspects like disaster recovery. Putting versions of your application in another location is beneficial for that. It’s also useful for provisioning applications that are not versatile by themselves, but there is a need for them. The real driving factor has to do with a push to agility and shrinking lead times. As people become more comfortable with the cloud and see people apply the technology successfully, we’ll gain confidence and push it to the big dynamic projects.

What are the obstacles? This is the distributed computing problem all over again. There are eight things that people assume when they write their software which turn into problems:

no security threat,
a homogenous environment,
single administrative domain,
zero latency,
infinite bandwidth,
fixed apology,
the network is free and
continuous connection assumptions.

If you’re going into the cloud, all of these start to become much bigger factors. You have to test them and account for them. Compliance and certification is an obstacle that will evolve and change. Licensing restrictions coming from companies that were not built for the cloud. People have a problem with inertia, they are used to being able to touch and feel, and there is a “fear factor” involved with not knowing that. IT process management continues to be a challenge. Incidents, change management, etc. suddenly are affected by the cloud. Security and privacy is also a concern. Most of all, it’s familiarity, this is still new.

There is a big split between component-centric clouds and service-centric clouds. Are enterprises only after “legacy” component-centric clouds, or will they eventually want to rebuild their applications for scalable services like SimpleDB, BigTable, and so on? The question is who controls it. How much control do you have to have? It’s important at some point to get to real world examples and beyond diagrams. Enterprises are not using Google App Engine. You can’t take a lot of applications and move them to Google App Engine. If you need extra processing on the side, you just can’t do it there. People need the ability to control on a finer grain level.

The challenge is that open and portable doesn’t give you access to build on top of other platforms. Do we need strategic cloud partners or will vendors be providers for each vertical? We’ll see efforts to make the clouds more interoperable. In the future we’ll see more applications built that are able to span clouds. We’ll also see more integration and a better level of mass data management.

Is there more management and transparency on the horizon? We should not try to hide infrastructure from users. You can view everything from components to applications. If something isn’t working right, what do you do? If there’s no transparency it will be hard to address. Amazon has done a good job on transparency. A lot comes down to end-to-end response time. What’s the business we’re supporting? What are the metrics and visibility required to say move 5 terabytes of data within the cloud? One of the challenges for large enterprises is the service component. Inefficient code means that money and time is lost.

From a testing point of view, now you’re dealing with an elastic environment. You get a number of changes in testing, now you have things that are running - you can’t bring something like Google in-house. If you have a component version you can test components or do some scaling, dump them in the cloud and see what happens. You need some way to deal with the testing and deal with the business metrics. Will one cloud give better performance? Will one be more stable? And which is more important? While cost may be an issue for some, many people using the cloud are actually finding costs savings. How do I plan for performance? How do I do my capacity planning? Testing is a big part of this, and each trial is really expensive. In the cloud, it’s much simpler. You can just try. Don’t be so concerned with the specifics of capacity planning, think about the resources and how to scale it.

What happens to software licensing when moving to on demand infrastructure? There is a stack of infrastructure, operating systems, and applications. What’s the impact for a company like Microsoft getting into the cloud business? The challenge there is that Microsoft is not really known for low pricing and they do charge for their OS. The biggest cloud out there is open source. Microsoft will broaden the whole cloud market, but may not be able to undercut what’s already there. Microsoft has a big advantage if they decide to do something with OS7 or something else, where they can offer something already virtualized. It would literally provide virtualized environments everywhere for everyone.

From a testing point of view, when we’re building applications that are distributed, doesn’t this put a lot more burden on the tester? The cloud can create additional testing requirements, especially regarding security. Part of the problem ends up being that when you involve more than one vendor, everyone points fingers for blame. Now you’re adding two or three more layers to the same thing. What makes the cloud different from virtualization? Virtualization is flexibility, the cloud wants to know how to manipulate. Testscripts are going to eventually need to be automated.

Highlights: