Web-Based Testing and Capacity Planning - High-Performance Web Databases: Design, Development, and Deployment - page 632

an automated software quality tool to expedite repetitive testing tasks.
Additionally, these test tools will enable your organization to perform
stress testing, which is key to ensuring sufficient network and server
resource levels for the production environment.
Capacity Planning
Effective performance testing is difficult without an accurate depiction
of future loads. Many companies simply over-engineer hardware and net-
works at high costs to minimize potential performance issues leading to
service degradations or deal with performance issues on a reactive basis.
Reacting to performance issues in today's highly competitive marketplace
could ultimately lead to the loss of customers during system downtime or
periods of poor performance. Planning capacity is a critical step required
to ensure the future performance of your web-based environment. The key
components involved are network, server (e.g. memory, CPU, I/O) and stor-
age capacity.
Establishing performance benchmarks and subsequently estimating
future growth is critical to planning the capacity of the network and serv-
ers. Although benchmarks are published by the Standard Performance
Evaluation Corporation for web servers (www.specbench.org), there uses
are limited and do not accurately represent a real-world integrated web
environment. Alternatively, benchmarks can be determined through stress
testing and mapping of performance (e.g. response times) to specific net-
work or hardware configurations under varying loads. Modeling tools and
techniques can also be used to determine performance characteristics
under varying loads.
Once initial benchmarks are established, future production loads can be
estimated using historical growth statistics and/or growth predictions esti-
mated by various Internet-analyst groups (e.g. IDC, GartnerGroup, and For-
rester Research). Subsequently, the growth forecasts can be put to test to
determine the resource and scalability requirements of the network and
hardware in the future. Note that peak loads of 3-4 times average loads
should be tested during the stress test phase. Additional stress testing con-
siderations is to model higher-volume loads for cyclical periods. For exam-
ple, online retailers may have much higher loads during the Christmas
period than during the rest of the year. Ensuring performance, especially
during these peak periods, will have an impact on web-site success. For
this reason, over-provisioning hardware or network components to a cer-
tain level is justified.
Although effective capacity planning should enable your systems to
handle future growth, monitoring of your networks and server resources
should continue to ensure capacity is within acceptable limits.