The Unanswered Questions Of Cloud Technology

Everyone is out there selling the cloud like there’s no tomorrow, so it’s tempting to think the technology is all done. But there are people out there who say that is not the case.

The cloud is still scaling up, and it’s not always a hundred percent clear how services will be operated, as more and more users turn to public shared resources for more of their IT.

The dream of cloud activists seems to be a shift away from in-house IT , and long term outsourcing deals, running on stable hardware in known locations. Instead, resource locations will be flexible, and customer processes will be running in an increasingly ad hoc, transitory way, on different providers’ services.

The new cloud deal

That will make new demands on service providers. The fresh requirements will be both technological and contractual – and where possible cloud providers will use technological means, to automate and re-automate the services they offer to meet customer needs

“Service level management will be crucial,” says  John Manley, director of cloud services at HP Labs in Bristol, UK, who spoke to TechWeekEurope after a keynote at  last week’s IP Expo in London.

For him, a full implementation of the vision of the cloud would involve allowing your service provider to subcontract parts of the offering to other providers if they can do it cheaper.  That may seem more radical than some cloud visions, but it’s emerged from the heritage of HP Labs, which has been experimenting with cloud services since before the name was coined – back in the 1990s, these were known as ASPs (application service providers).

Later, in the middle-2000s, Manley and the Labs spoke of “Grid” computing and offered shared resources for high-performance computing, notably a platform where film companies could buy batch rendering services.

The idea here was to offer a spot market, where the price of a batch rendering job could vary in real time, and customers could choose which provider they went with for a particular piece of work.

This sort of advanced cloud market produces particular problems, and Manley suggests two which are at the front of the minds of HP Labs: Federated SLA Management, and Multitenancy,

Federating your SLAs

To get the full flexibility of a cloud market, cloud providers should be able to sub-contract work, he says, but that’s only practical if they can guarantee that the provider they hand the work on to will meet the service requirements of the original customer.

So when a customer books a job, they will ask for a particular level of performance, and make a service level agreement. The cloud provider will have to enforce at least that level of service from any sub-contractors.

Doing this manually would be impractical, and introduce overheads enough to destroy the benefits of the cloud, so it’s got to be automated. HP Labs is beavering away, therefore on “Federated SLA Management”, a concept which Manley thinks will usher in greater flexibility and cheaper, more usable clouds.

It’s got to work well, and it’s got to take into account geo-politics, so jobs only get moved to a different provider if the end customer has approved that any personal data can  go there – or if it is hashed and anonymised.

It also has a fundamental requirement for a communicable standard for SLA agreements – something which should emerge in XML form, and allow the customer to be sure that any ad hoc partners meet their needs.

Multple tenants in an instance

The other technical problem comes from the idea that the jobs in the cloud will steadily get more granular. If animation companies really are sending small bits of work to rendering houses, almost at random, they will need to know their data is kept private.

Pixar and Dreamworks obviously have their own rendering facilities, but imagine two smaller studios, sending scenes out to a rendering house on the cloud. The first one would want to be absolutely certain that the second could not see the material it was working on.

This is easy to ensure if the two use separate resources, and that can be guaranteed if fairly large jobs are being done, and each gets its own specific resources.

However, Manley foresees a world where there would be so many little rendering jobs coming and going, that the cloud provider could not launch separate instances of the rendering engine for each one. “If you have 1000 customers, can you really launch 1000 instances of the software?” he asks. “Eventually it becomes unaffordable, and you have to have multitenancy in the software!”.

Will users accept this? Well, I’ve never thought about it specifically, but I would bet that a service like Google Drive is not firing up a separate instance of a word processor for every document being edited.

At the level of casual writing and journalism, I sort of trust that. If I were writing something I believed anyone would actually want to steal (or even read) I’d have to think carefully and ask more questions.

Multienancy is being looked at in various places, and so is Federated SLA management – so if it became a worry, I might have to wait till Manley and his crew or some other team comes up with an answer.

Do you know about Windows? Try our quiz!