William Vanbenepe looks at having 'standard' cloud APIs, following on from a Forbes article.. Hmmm.

Some quick thoughts

There's management APIs, and there is the architectures the apps will be built with. If you look at the past grid work at GGF, nobody differentiated these, and the APIs were driven from the needs of organisations trying to manage the grid, rather than the needs of the applications. So we'd be in these meetings discussing advanced WS-ResourceFramework issues, but the general use case for all of this was Tomcat serving up HTTP. Oh the irony.

The world is moving towards REST. Its the only thing that's been shown to work over space and time. So anything that lives in WS-* has to be viewed as transient.

It will be a long time -if ever- before the services for apps hosted 'in the cloud' stabilise, and hence the APIs. Take the database layer, for example. Apache HBase != Amazon SimpleDB ! Apache CouchDB != MS SQL Server for the cloud != whatever MySQL/Sun will surprise us with one day. To try and standardise that would be a mistake. All we need to do know is agree that REST is the right long-haul model

You don't need standards with an open source implementation -the open source implementation will do on its own, just fine. For example, Hadoop is a good API for Java code to use if it wants to work with multi-PB filesystems. Its not "standard"; it isn't even stable -it's evolving on a monthly basis. But if you want to run data-local algorithms like MapReduce, it's there and it works. I don't see how anything like a JCP Map-Reduce API working group would be be helpful.

The EC2 API is not that admirable; cloning that API is wrong. You have to talk to the servers using SOAP or HTTP Query API, asking for a number of machines; you get a list back. To make sure they are there, you need to keep polling/pinging them. Why can't I push out a model that lists what I want, including network, security, apps on the machines, etc, and have the infrastructure keep that configuration up until I push out a new one?

There was a little event run by HPLabs and MSResearch recently, on Declarative Datacentres. A lot of the stuff in the position papers is pretty interesting. Read the proceedings