Capacity Planning - The Practice of Cloud System Administration - page 386

18.6 Summary
Capacity planning is the process that ensures services have enough resources when they
are needed. It is challenging to prevent service interruptions due to lack of capacity and,
simultaneously, preserve capital by adding only the capacity required at any given time.
Standard capacity planning is based on current usage and simple rates of change. It as-
sumes future resource needs will be similar to current usage plus two kinds of growth.
Normal or organic growth is what is expected based on current trends. Planned growth is
whatisexpectedduetonewinitiativessuchasmarketingplans.Additionalcapacity,called
headroom, is added to handle short-term spikes. Based on the timetable showing lead time
(i.e., how long it takes to acquire and configure new resources), capacity schedules can be
determined. By reducing lead time, capacity planning can be more agile.
Standard capacity planing is sufficient for small sites, sites that grow slowly, and sites
with simple needs. It is insufficient for large, rapidly growing sites. They require more ad-
vanced techniques.
Advanced capacity planning is based on core drivers, capacity limits of individual re-
sources, and sophisticated data analysis such as correlation, regression analysis, and stat-
istical models for forecasting. Regression analysis finds correlations between core drivers
and resources. Forecasting uses past data to predict future needs.
With sufficiently large sites, capacity planning is a full-time job, often done by project
managerswithtechnicalbackgrounds.Someorganizationsemployfull-timestatisticiansto
build complex models and dashboards that provide the information required by a project
manager.
Good capacity planning models can also detect unexpected changes in resource
needs—for example, a new software release that unexpectedly requires more resources.
Capacity planning is highly data-driven and uses past data to predict future needs.
Launching a brand-new service, therefore, poses a special challenge. Dark launches and
other techniques permit services to gather accurate data before the service is visible to
users.
To improve cost-effectiveness, reduce the time it takes to provision new resources. Pro-
visioninginvolvesacquiring,configuring,andputtingnewresourcesintoproduction.Total
provisioning time is called lead time. Long lead times tie up capital. Reducing lead time
reduces idle capacity and improves financial efficiency.
Capacity planning is a complex and important part of reliable operations.