2.3System Configuration and Virtualization

How a system is configured is a key design decision that determines the balance of performance, reliability, scalability, and cost. After understanding the advantages and challenges of concentrating processing on one machine versus distributing it across many, we cover virtualization and the cloud as modern ways to realize these configurations.

2.3.1Centralized/distributed processing and client-server

• Centralized processing concentrates processing on one high-performance central computer. Management is easy to unify and control, but that single machine can become a single point of failure that halts the entire system if it goes down. Distributed processing spreads processing across multiple computers. Even if one fails, the others can keep running, and load can be shared—but the configuration becomes more complex and management cost rises.
• The client-server system is a representative form of distributed processing that splits roles between a client, which requests processing, and a server, which fulfills the request by providing a service. The three-tier web architecture separates a web system into a presentation tier (screen rendering, web server), an application tier (business logic, app server), and a data tier (data storage, DB server), allowing each tier to be scaled up and maintained independently.

2.3.2Clustering and virtualization

• Clustering bundles multiple servers so they behave as if they were one. It broadly splits into clusters aimed at failover (taking over processing when a failure occurs) and clusters aimed at load balancing (sharing processing across multiple machines). Load balancing is the mechanism by which a load balancer distributes requests across multiple servers, preventing load from concentrating on any single one.
• Virtualization is a technology that splits and shares a single physical machine's hardware resources across multiple virtual machines (VMs). Its core component is the hypervisor, which comes in a bare-metal type (Type 1) that runs directly on physical hardware, and a hosted type (Type 2) that runs on top of an existing OS. Virtualization lets one physical server run multiple OSes and applications independently, raising consolidation ratios.

Consider planning the infrastructure for a new web service. Even if you start with a centralized configuration—one server hosting the web server, application, and database together—as usage grows you run into problems: "a traffic spike bogs down the whole server," or "fixing one feature requires taking everything down." Migrating to the three-tier web architecture, separating the presentation, application, and data tiers, lets you handle each tier independently—for example, scaling up only the application tier, or hardening backups only for the data tier. Going further, you can add more application-tier servers and place a load balancer in front to form a load-balancing cluster, spreading load per machine while keeping the service running on the remaining servers if one fails. You also need to choose an implementation approach for the infrastructure itself. Provisioning each server as a dedicated physical machine takes time to procure and yields low consolidation, so virtualization (multiple VMs on a hypervisor) or the lighter-weight container approach (which shares the OS kernel and packages only the application and its dependencies; it starts faster and consumes fewer resources than a VM) lets even one physical server flexibly host many execution environments. If you also want to outsource procuring and operating the infrastructure itself, you consider the cloud: IaaS, where you rent only infrastructure like virtual machines and basic networking and manage the OS and above yourself; PaaS, where the application runtime and middleware are also provided so you can focus purely on implementing the application; and SaaS, where the finished business application itself is provided and you only use and configure it. Choosing among these based on how much you want to manage yourself versus hand off is the practical judgment.

2.3.3Section summary

• Centralized processing tends toward a single point of failure; distributed processing gains availability at the cost of management complexity. The three-tier web architecture lets each tier be scaled and maintained independently
• Clustering serves either failover or load balancing. Virtualization (hypervisor) comes as bare-metal (Type 1) or hosted (Type 2). Containers share the OS kernel and are lighter than VMs
• IaaS = user manages OS and above; PaaS = user manages only the application; SaaS = only usage/configuration (managed scope, narrowest to broadest: SaaS < PaaS < IaaS)