In the Cloud Conceptual Reference Diagram provided by NIST, we can distinguish Cloud consumers, Cloud providers and Cloud Service Management for business support.
In virtualization, necessity is the mother of invention. Josh Simons saw enterprise trends including Google, Yahoo, Amazone.com and Facebook.
Convergence is driven by increasingly shared concerns consisting in scale-out management, dynamic workloads, high utilization, multi-tenancy and security, power management, low latency communication, and application parallelism.
Today, we are confronted with data intensive computing and big data anaytics. MapReduce could be a solution.
If we look below the Clouds, there is virtualization for HPC. Here the following requirements are essential, according to Josh Simons. Virtualization for HPC has to be application fault tolerant. There is a need for
checkpointing on the reactive side and predictivity on the proactive side.
A dynamic resource management has to move running jobs for efficiency, resiliency, power management, preventative maintenance, etc.
In heterogeneous environments, you should be able to run your own software stack, including OS, libraries, and applications, and not the site's stack.
Workload isolation is important too for secure multi-tenancy, failure protection, and even performance. And for the current virtualization users there should be a unification of the IT infrastructure.