The technical challenges holding back cloud computing

While companies try to solve cloud computing issues the market remains some way off maturity, writes Danny Bradbury

While companies try to solve cloud computing issues the market remains some way off maturity, writes Danny Bradbury.

Gartner recently suggested that cloud computing as used for service-enabled applications still had seven years until it reached market maturity. Some of the problems it faces include scalability, interoperability, and security, not to mention more business-focused issues.

For example, what is the most sensible way to pay for these services? Companies are piling considerable resources into research and development to solve some of these issues. HP, Intel and Yahoo created a research effort last July, partnering with several academic institutions around the world.

And the European Union has also been involved in driving forward some research designed to answer questions such as how clouds can be made more interoperable.

John Manley, director of the automated infrastructure laboratory at HP labs, pinpoints scalability as a particularly challenging issue.

Conventional environment

In a conventional multi-tenant hosted environment, you might simply use a copy of an application for each user, to provide the necessary isolation that customers needed. "Imagine that this becomes a popular service with a million users. Can you imagine having a million instances of that service?" he asks.

"You now have to start thinking about being multi-tenant not in the infrastructure, but at the application level. Now, all the isolation has to be built into the application, and that is a much trickier thing to do."

Large-scale SaaS companies have been dealing with such problems commercially, which will help to inform research and development in this area. But the industry has to develop to the point where this becomes a natural way of doing things, he adds. The cloud isn't singular, of course.

We already have different ones from a variety of vendors. As they become more commercially appealing, we must avoid the vendor lock-in that we saw with the original 'cloud' - the mainframe.

"We have a set of isolated clouds at the moment, and we are a long way off having a holistic, overarching cloud out there that I can load my data into anywhere I want and access it from anywhere I want," says Alan Priestley, strategic marketing manager at Intel.

There are some moves afoot to try to sort this problem. For example, the European Union has shelled out €17m on an IBM-led project called Resources and Services Virtualisation without Barriers (RESERVOIR). The initiative, led by IBM, aims to enable IT services to be moved between different machines in the cloud to accommodate fail-over processes, and workload management.

Remotely mirrored servers

We have already seen this to some extent, with conventional remotely mirrored servers, for example. But they don't have anything like the same flexibility. Solutions such as VMware's VMotion move virtual machines around between physical devices automatically, but try moving those virtual machines between different companies' hypervisors in remotely connected datacentres.

"If you have a standard communication protocol across these two clouds, where you can then communicate with the handlers of the other cloud, for the right security, networking, availability, and orchestration, only then can you make it happen," says Radha Ratnaparkhi, director of commercial systems at IBM. In early March, the company demonstrated live migration of SAP applications across remote IBM systems.

These technical challenges may be daunting enough, but they are also intimately linked to more business-focused issues. Dr Marcel Kunze works at the Steinbuch Centre for Computing, one of six research centres of excellence around the world that are supporting the HP, Intel, and Yahoo research effort.

Once you have figured out how to move the data around effectively, you have to make sure that you can stop it from moving just anywhere, he points out. "It may be important for policy reasons that data has to stay in Europe and not move to the US, for example," he says. "There are no methods available at the moment to define the location of that storage."

How do you pay for it?

Then when all that has been worked out, we have to consider the pure economics of cloud computing. How do you pay for it? Ratnaparkhi cites several potential models, including pay as you go, and a model that enables users to bid for time in the cloud. HP has already run trials in which people wanting to render animations could bid for time in this way.

Ultimately, we hope that the cloud will become a more ubiquitous part of everyday computing life. But we still have some way to go before the market as a whole latches on to it. Creating confidence among enterprise users requires some hefty research work first. While the white coats labour in the labs, we'll wait patiently, and continue to dabble.

Read more on Networking hardware