Cloud technologies have reached levels of maturity and market penetration such that organizations have to justify why they are not using cloud rather than the other way around.
This has occurred because cloud technologies offer real benefits to users and organizations. With cloud infrastructures, people can easily deploy large applications and can dynamically tune the allocated resources to maximize responsiveness and reliability. This flexibility has made the use of cloud technologies economically attractive by reducing capital and/or personnel costs.
There are three critical components of cloud technologies that have lead to its success:
- Mature Virtualization Technologies
- Virtualization techniques have been directly integrated into mainstream operating systems and modern CPUs. This integration allows use of virtualization with only a negligible impact on performance.
- Simple, Universal APIs
- Most clouds use Resource Oriented Architectures and “REST” APIs over the ubiquitous HTTP protocol. Doing so makes the service universally accessible from all programming languages and reuses the well understood HTTP service model.
- Ubiquitous, Reliable Networking
- Robust, high-bandwidth cellular networks, wifi, and wired Internet connections provide universal, reliable, 24/7 access to critical services hosted in remote infrastructures like clouds.
By combining these components, cloud platforms provide efficient, powerful computing resources that consumers can easily and reliably access from anywhere.
The constant barrage of marketing using the term “cloud” can make it difficult to develop a precise understanding of what cloud technologies are. Fortunately, the American standards institute (NIST) provides clear, precise definitions that fit our needs. These definitions are the de facto standard for discussing cloud platforms.
What resources (or services) are provided by the cloud? In answer to this question, NIST and others define three “service models”:
- Software as a Service (SaaS)
- Provides a complete application to a customer hosted on a cloud platform to provide lower latencies, better bandwidth, scaling, or other features. Typically the customer will access the service through a web browser or another client on the customers computer.
- Platform as a Service (PaaS)
- Provides a programming environment and cloud infrastructure featuring high-level capabilities like load-balancing, scaling, etc., relieving the programmer from having to construct those services from scratch in the application. Typically the customer is an application developer who accesses the service through a proprietary, language-specific API.
- Infrastucture as a Service (IaaS)
- Provides access to raw computing resources (virtual machines, storage, etc.) that can be provisioned (and released) rapidly. Customers access these services either through a simple (usually REST) API or through a web interface.
These service models are often presented as a hierarchy as a PaaS is often built over a IaaS, as well as a SaaS over a PaaS.
Despite these clear-cut definitions, real services tend to be more complicated, offering elements of the different service models from the same cloud infrastructure.
The “deployment model” answers the question: Who uses the cloud infrastructure? NIST defines three deployment models:
- These are infrastructures in which the computing resources are co-located with its primary users. The users (or their institute) buy the computing resources directly and run them as a cloud for their own purposes.
- These are infrastructures run by and for a group of collaborating institutes with similar aims. The users usually buy some fraction of the computing resources and share those resources with others in the community. Allocation of resources between people is usually done via “horse trading.”
- Cloud infrastructures that offer their resources to the general public. The customers pay for use of the computing resources directly (usually via a credit card). The computing resources are housed in data centers controlled by the owner of the cloud, not by the customer or the customer’s institute.
NIST actually defines a fourth deployment model, Hybrid Cloud, which is really just a mix of the other deployment models. This usually comes up in the context of “cloud bursting”, where remote cloud resources (public clouds) are used when a local cloud resource (private cloud) becomes saturated.