Cloud Computing Unit 3: Cloud Architecture, Services and Storage

A cloud is easier to understand as a stack of four layers. Each layer uses the one beneath it and offers something more useful to the one above.

At the bottom, the datacenter layer is the physical hardware: servers, routers, switches, power supply and cooling. It supplies raw CPU, memory, disk and bandwidth. Above it, the infrastructure layer uses virtualization software such as Xen, KVM or VMware to split those physical resources into virtual machines and virtual storage that can be handed out on demand. This is the core of a cloud: resources are provisioned dynamically instead of being tied to one machine.

The platform layer runs on those virtual machines and provides an operating system and software frameworks, so developers can deploy applications without setting up servers. At the top, the application layer is what cloud users actually touch: web services, multimedia and business applications delivered as they need them. The top three layers line up with the three service models later in the unit: SaaS, PaaS and IaaS.

• Application layer: cloud services used by end users, such as web services, multimedia and business applications.
• Platform layer: operating system and software frameworks for deploying applications on virtual machines.
• Infrastructure layer: virtualization (Xen, KVM, VMware) turns hardware into on-demand virtual machines and storage.
• Datacenter layer: physical servers, switches, routers, power supply and cooling.

In a public cloud, a provider makes infrastructure and computing resources available to the general public over a public network such as the internet. Many customers share the same pool of hardware, and each uses only what it needs. Enterprises often move commodity workloads to public providers instead of running the hardware themselves.

Two common misunderstandings are worth clearing up for the exam. "Public" does not mean free: a public cloud can be free or inexpensive, but usually you pay for what you use. And "public" does not mean your data is visible to everyone: providers give each customer access controls.

Typical public cloud workloads include public-facing websites, wikis and blogs, batch-processing jobs and data-intensive workloads. Public clouds offer resource pooling, self-service, usage accounting, elasticity and multi-tenancy.

• Advantages: flexible, reliable, highly scalable, low cost and usable from anywhere.
• Limitations: less control over security and limited customization.

A private cloud is deployed for one organization, traditionally inside the company's firewall on servers it controls. It keeps cloud-style benefits such as on-demand capacity, self-service provisioning and service-based access, while giving the organization tighter control over security, resilience and who is allowed in.

The trade-off is cost and scale. Because the organization owns the hardware, it loses true pay-per-use pricing, has to size its hardware for peak load, and carries the full operating cost. A private cloud makes sense when control and security requirements justify that cost.

The notes describe three variants: a shared private cloud, where business units share capacity and are charged by usage; a dedicated private cloud, with its own catalogue of services provisioned on demand; and a dynamic private cloud, where workloads can move between shared and dedicated resources.

• Advantages: high privacy and security, and more control.
• Limitations: limited scalability, higher cost and restricted access by design.

A hybrid cloud combines public and private clouds and uses them at the same time. A common pattern is to keep critical applications and sensitive data in the private cloud, while sending non-critical applications, or extra computing demand, to a public cloud.

Many organizations use a hybrid cloud as an intermediate step: an on-ramp from their existing IT environment into the cloud. It aims to combine the scale and convenience of a public cloud with the control and reliability of a private one. The lesson's diagram also shows a community cloud, which several organizations with shared requirements use together.

• Advantages: flexible, keeps critical data secure, cost-effective and scalable.
• Limitations: more complex to run, and security compliance has to be managed across both environments.

Cloud providers sell their layers as services. The difference between the three service models is how much the provider manages for you.

Infrastructure as a Service (IaaS), sometimes called Hardware as a Service, lets an organization outsource storage, servers and networking. You rent virtual machines and pay per use, often billed hourly, daily or monthly, and you still install and manage the operating system and applications. Capacity can be changed within minutes instead of buying hardware. Amazon EC2 and Amazon S3 are well-known examples.

Platform as a Service (PaaS) provides a ready platform to develop, run and manage applications: operating system, server software, database, storage and development tools. You deploy your code and the provider handles the infrastructure underneath. Examples include Google App Engine and Microsoft Azure. The main risk is getting locked into one provider's platform.

Software as a Service (SaaS) is a complete application hosted on the provider's servers and used over the internet, usually in a browser. Web email such as Gmail is the classic example. SaaS comes in two broad categories: line-of-business services sold to companies on subscription, such as CRM and supply-chain tools, and customer-oriented services offered to the public, sometimes free and supported by advertising.

The notes also list the problems architects have to plan for when designing a cloud. They are worth knowing as a short-answer list.

• Service availability and data lock-in.
• Data privacy and security concerns.
• Unpredictable performance and bottlenecks.
• Distributed storage and widespread software bugs.
• Scalability, flexibility and standardization.
• Software licensing and reputation sharing.

Cloud storage means keeping data online in the cloud, where it is stored on, and accessible from, many distributed and connected resources. It offers better accessibility and reliability, quick deployment, strong protection for backup, archiving and disaster recovery, and lower overall cost, because you don't buy and maintain your own storage hardware.

There are four types. Personal cloud storage, also called mobile cloud storage, syncs and shares one person's data across devices; Apple iCloud is an example. In public cloud storage, the enterprise and the storage provider are separate and the provider fully manages the storage. In private cloud storage, the provider's infrastructure is integrated into the enterprise's own data center, which helps with security and performance concerns. Hybrid cloud storage combines public and private storage.

• Advantages: easy to use and reach from anywhere, built-in disaster recovery and lower cost.
• Limitations: needs a reliable internet connection and enough bandwidth, and you depend on the provider for data security.

Cloud providers mainly use two charging models. In the utility model you pay per use, much like an electricity bill: the more compute or storage you consume, the more you pay. In the subscription model you pay a time-based fee, for example every month, for access to a service.

The lesson ends with the challenges providers themselves face: protecting against data corruption, keeping logical security strong, making data work across different systems, and handling software vulnerabilities and breaches.

• Utility model: pay per use, like electricity charges.
• Subscription model: time-based charges for using a service.