Appendix A. Benchmarking 479
Caution using benchmark results to design production
While taking benchmark results as the foundation to build your production environment
infrastructure, you must take a close look at the benchmark performance information and
watch out for different points:
Benchmark hardware resources (servers, storage systems, SAN equipment and network)
are dedicated only for this performance test.
Benchmark infrastructure configuration must be fully documented. Ensure there is no
dissimilar configurations (for example, dissimilar cache sizes, different disk
capacities/speed).
Benchmarks are focused on the core application performance and do not often consider
interferences with other applications that can occur in the real infrastructure.
The benchmark configuration must be realistic (technical choice regarding performance
versus usability or availability).
Scenarios built must be representative in different ways:
– Volume of data
– Extreme or unrealistic workload
– Timing execution
– Workload ramp-up
–Avoid
side effects such as populating cache, which could generate incorrect results
Be sure to note detailed optimization actions performed on each of the components of the
infrastructure (including application, servers, storage systems, etc.).
Servers and storage performance reports must be fully detailed including bandwidth, I/O
per second and also response time.
Understand the availability level of the solution tested, considering the impact of each
component failure (HBA, switch, RAID, etc.).
Balance the performance results with your high availability requirements. First, consider
the advanced copy services available on the specific storage system to duplicate or mirror
your production data within the subsystem or to a remote backup site. Then, watch out
that performance is not equivalent if your data is mirrored or duplicated.