Cloud Operating System – what is it really?


A recent article published on, “Are Cloud Operating Systems the Next Big Thing”, suggests that a Cloud Operating System should simplify the Application stack. The idea being that the language runtime is executed directly on the hypervisor without an Operating System Kernel.

Other approaches for cloud operating systems are focussed on optimising Operating System distributions for the cloud with automation in mind. The concepts of IaaS (Infrastructure as a Service), PaaS (Platform as a Service) and SaaS (Software as a Service) remain in the realm of conventional computing paradigms. 

None of these approaches address the core benefits of the cloud. The cloud is a pool of resources, not just another “single” computer. When we think of a computer, it has a processor, persistent storage and memory. A conventional operating system exposes compute resources based on these physical limitations of a single computer. 

There are numerous strategies to create the illusion of a larger compute platform, such as load balancing to a cluster of compute nodes. Load balancing is most commonly performed at a network level with applications or operating systems having limited exposure of the overall compute platform. This means an application cannot determine the available compute resources and scale the cloud accordingly.

To fully embrace the cloud concept a platform is required that can automatically scale application components with additional cloud compute resources. Amazon and Google both have solutions that provide some of these capabilities, however internal Enterprise solutions are somewhat limited. Many organisations embrace the benefits of a hosted cloud within the mega data centres around the world. Many companies have a requirement to host applications internally.

As network speeds increase the feasibility of a real “Cloud Operating System” becomes a reality. This is where an application can start a thread that executes not on a separate processor core, but executes somewhere within the cloud. 

A complete paradigm shift is required to comprehend the possibilities of an Operating System providing distributed parallel processing. Virtualisation takes this new cloud paradigm to a different level where the abstraction of the hardware using a virtualisation layer and a platform operating system presents compute resources to a Cloud Operating System.

The same way as a conventional operating system determines which CPU core is the most appropriate to execute a specific process or thread, a cloud operating system should identify which instance of the cloud execution component is most appropriate to execute a task. 

A cloud operating system with multiple execute instances on numerous hosts can schedule tasks based on the available resources of an execute instance. By abstracting task scheduling to a higher layer the underlying operating system is still required to optimise performance  using techniques such as Symmetric Multiprocessing (SMP), processor affinity and thread priorities.

The application developer has for many years been abstracted from the hardware with development environments such as C#, Java and even PHP. Operating systems have not adapted to the Cloud concept of providing compute resources beyond a single computer. 

The most comparable implementation is the route taken by Application Servers with solutions such as JAVA EJB where lookups can occur to find providers.  Automatic scalability is however limited with these solutions.

Hardware vendors are moving ahead by creating cloud optimised platforms. The concept is that many smaller platforms create optimal compute capacity. HP seem to be leading this sector with their Moonshot solution. The question however remains: How do you make many look like one?  

Enterprises have existing data centres where very little of the overall compute capacity is actually leveraged on an ongoing basis. When one system is busy, numerous others are idle. A cloud compute environment that can automatically scale across a collection of servers providing actual cost savings. Compute capacity would be additive using existing infrastructure for workload based on available resources. According to the IDC report on world wide server shipments the server market is in excess of $12B per quarter. The major vendors are looking for ways to differentiate their solutions and provide optimal value to customers.

Combining hardware, virtualisation and a Cloud Operating System organisations will benefit from a reduction in the cost to provide adequate compute capacity to serve business needs.

Gideon Serfontein is a co-founder of the Bongi Cloud Operating System research project. Additional information at