Multiple users or groups of users will be assigned fine-grained permissions on what types of servers and applications they may access, based on management policies that assign the permissions automatically.
User interactions can be governed by a strictly limited menu of options, or by offering the end user full, programmatic control over a server cluster, depending on how the service provider wishes to operate, said Pinkham. Amazon Web Services sticks to the menu of options approach as the best way to offer compute cycles cheaply. A scientific researcher or technology inventor might want programmatic control where he's submitting programming of his own choosing, assessing runs of the program and then submitting changes -- in a way that couldn't be governed by a fixed menu of options.
Feedback from early implementers will be one measure of how well Director works with a variety of storage management systems. Pinkham said Director has been engineered to work with some of the most widely installed ones.
A cloud is not just a set of virtualized servers or a typical data center set up to be publicly accessible over the Internet, said Nimbula's VP of sales Martin Buhr during a panel on open source code in private cloud computing. "It changes in some major ways the way you write applications. The best practices on Amazon should become best practices in general," he said.
One of the main practices pioneered by Amazon.com, Google, and others is the movement of fault tolerance from a hardware solution, once represented by the paired servers of a Tandem Computers system, into a software solution, where the cloud software knows how to compensate when a server fails. Most NoSQL data management systems, such as CouchDB or Cassandra, have adopted the practice.
Willem van Biljon, Nimbula's VP of products, generalized that conclusion further as a speaker at the event's general session Wednesday. "The lessons from large scale computing (such as Amazon's EC2) matter at all scales," he said. There is no one size of enterprise or data center that should adopt the tenets of cloud computing; they all should, he suggested in his compact presentation.
Problems still be solved before cloud computing can be adopted at many levels include automating the provisioning of network security when commissioning virtual machines and assigning them addresses on the network. In addition, "identity is a problem" as users try to make use of both the on-premises data center and enterprise workloads in the public cloud. "If we can figure out how to federate identity -- that's one of the keys to the problem. If we can solve that, we can get the rest right," he predicted.
In listening to these three representatives of Nimbula across three days, one comes away with the impression that it is a company working on solving those problems and intent on applying the experience gained in building EC2.
Director is due to become available in early April and will be free for use in implementations up to 40 cores (which in some cases, would be a cluster of five, 8-core Intel Nehalem servers). Nimbula will charge for Director above 40-core implementations, along with furnishing technical support contracts.