Why Platform-as-a-Service is the Operating System Of The Cloud

When I took my operating systems fundamentals course in college I was taught that an operating system provides very specific capabilities that provides users with access compute resources for building and running applications. Over time as networking capabilities and bandwidth increased, the notion of a set of modules that interface between the user and the hardware has changed to incorporate concepts of distributed operating systems, network operating systems and autonomous systems. While the notion of the operating system may have changed certain attributes have remained constant:

• scheduling processes
• coordinating interaction among processes, interprocess communication and synchronization
• managing system resources
• enforcing access control and protection
• maintaining system integrity and performing error recovery

When looking at container-based PaaS offerings, such as CloudFoundry and Heroku, one can see many of these functions in operation across a set of virtual compute resources. If we consider that Infrastructure-as-a-Service (IaaS), bare metal and virtualized hardware inclusive of traditional operating systems, such as Windows and Linux, all represent the modern day equivalent of a compute node in a cloud universe then we can take the leap that the PaaS provides the interface between the user and that node. Moreover, we can include in this list of resources the services that support application operations, such as identity management, data management, messaging and monitoring.

If we explore the role of the PaaS in cloud application development and delivery, we can see that the platform overlays a set of cloud nodes and services exposing their resources to the application runtime environment. The PaaS then handles application lifecycle management inclusive of execution, process allocation and resource scheduling, access control and protection fostering multitenancy, and error recovery. Hence, container-based PaaS meets with the criteria to be considered an operating system.

Perhaps even more interesting is that the comparisons that can be drawn with regard to application development for a single operating systems versus a cloud operating system. One constant that remains as we look across time at operating systems is the increasing level of abstraction. Each level of abstraction has afforded us the ability to focus less on resource limitations, but even a cluster of virtualized compute resources still has capacity limitations.

With the emergence of a cloud operating system, we have the opportunity to finally escape those limitations by spanning and aggregating clusters of virtualized compute resources. Moreover, provisioning of these resources are delegated to services that are designed optimally for the physical resources they manage. For example, the cloud operating system/PaaS can communicate with the cloud management systems to identify where there resources that can satisfy the need for very-high speed (I/O per second) storage. Each cloud management system can then list its resources and corresponding metrics and availability. The PaaS can then request the one that best meets the criteria for the application. That is, the PaaS is the one environment that knows enough about the performance of the application and can schedule the resources and bind them to the application.

With these types of abilities, we can start our designs for cloud applications with the perspective of plenty instead of managing to the constrained. We can start to specify service levels within our applications, which can then be interpreted by the PaaS and turned into bound resources during execution. We have appropriate division of work across an application execution supply-chain. The bare metal provides maximum movement of bytes to the physical device. The hypervisor divides those resources into dynamically allocated blocks. The virtualization clusters allow those blocks to be moved around to maximize utilization of resources. And the PaaS can communicate across clusters to select the best set of available resources to ensure optimal execution of the application for which it is responsible.