3.4System Configuration Elements

Some systems run on a single computer; others use system configuration to combine multiple computers, each handling a specific role. The chosen configuration affects performance metrics (how fast the system responds) and reliability metrics (how well it resists failure). This section covers representative configuration patterns and the metrics used to evaluate them.

3.4.1System configuration: centralized, distributed, client-server, virtualization, and cloud

• Centralized processing concentrates processing on a single large computer (easy to manage, but the whole system stops if that one machine fails). Distributed processing spreads processing across multiple computers (load is shared, and the whole system is less likely to stop if one machine fails).
• A client-server system is a representative form of distributed processing that splits roles between the requester (the client) and the provider of the requested service (the server)—for example, a web browser and a web server.
• Virtualization uses software to create multiple computer environments on a single physical machine (running several systems while keeping the number of physical servers down). The cloud (cloud services) lets an organization use IT resources such as servers over the internet, on an as-needed basis, without owning the hardware itself.

3.4.2Performance and reliability metrics

• Response time measures the time from issuing a request until the first response begins to arrive (a measure of responsiveness). Throughput measures how much work can be processed per unit of time (a measure of processing capacity). A benchmark is a standardized test used to measure and compare system performance.
• Availability is the proportion of time a system is operating normally (e.g., 99% availability means the system runs normally roughly 99 out of every 100 hours). A higher value means a system that is less prone to downtime.
• RASIS is an acronym for five perspectives on reliability: Reliability, Availability, Serviceability, Integrity, and Security. MTBF (Mean Time Between Failures) is the average time a system keeps running before it fails—longer is better. MTTR (Mean Time To Repair) is the average time to restore a system after a failure—shorter is better.
• Fault tolerance is a design philosophy where the system keeps operating normally using the remaining equipment even if part of it fails. A dual system runs two fully independent systems simultaneously, so if one fails, the other alone continues processing. A duplex system keeps a normally used primary system and a standby (secondary) system that takes over when the primary fails.

Consider the backend system of an online shopping site to see how these metrics are used. When a shopper's smartphone app (the client) sends an order, the retailer's server receives and processes it—a typical client-server system. To handle traffic spikes during a big sale, instead of centralized processing on one large server, the company uses distributed processing across multiple servers, so that if one fails, the others take over and the whole site avoids going down. In recent years, rather than buying more physical servers, it has become common to provision several virtualized server environments on the cloud, only as needed. To evaluate this system's "speed," you measure the response time from pressing the order button to when the first response starts arriving, and the throughput (how many orders can be processed per second), using a standard benchmark test. To evaluate "resistance to downtime," you tally the MTBF (the average interval between failures, from operational history) and the MTTR (the average time from a failure to recovery), and compute the monthly availability from those figures. For a part of the system that absolutely must not stop, such as payment processing, a dual system—two fully independent systems running simultaneously—lets processing continue on the surviving system if one fails; for a somewhat less critical internal system, a duplex system of a normally used primary plus a standby that takes over on failure may be used instead. RASIS is the framework that brings together these five perspectives—reliability, availability, serviceability, integrity, and security—into one evaluation.

3.4.3Section summary

• Distributed processing, client-server, virtualization, and the cloud are the basic system-configuration patterns
• Performance is measured via response time and throughput, using a benchmark
• A longer MTBF and a shorter MTTR are both good. RASIS covers five reliability perspectives. Dual = two systems at once; duplex = primary + standby