Database Reference
In-Depth Information
In these types of benchmarks, it is the workload model itself that is used as the
reference, rather than the individual system configuration. Common industry-standard
database benchmarks include those produced by the Transaction Processing
Performance Council (TPC), such as TPC-C (OLTP), TPC-H (Ad-hoc Decision
Support), TPC-VMS (virtualized database benchmark), and others. Refer to
www.tpc.org
for a complete listing of benchmarks and their results.
Vendor Benchmarks
Throughout this topic, we have used the Dell DVD Store Benchmark as a tool to provide
consistent, repeatable performance test results. This is a good example of a vendor
benchmark and is a well-defined standard repeatable benchmark that is relatively quick
and easy to set up and use.
Another example of a vendor benchmark, which is useful for comparison, is the SAP SD
Benchmark. Although designed primarily to compare SAP ERP systems across all types
of different system architecture, including Unix and x86 systems as well as different
databases, it is a great OLTP workload and a well-defined standard test. You can find
SAP-certified benchmark results for Windows systems using Itanium, Intel, and AMD
processors and different versions of MS SQL Server. We discuss later in this chapter
how these benchmark results can be useful in comparing performance metrics from
different systems as part of your baseline and design process.
Developing Your Own Performance Benchmarks
Using industry-standard or vendor benchmarks can be useful to get a general idea of the
performance of your virtual infrastructure and design, as compared to other examples of
the same benchmark being run. However, this is not a valid representation of what you
should expect for your specific workloads when they are virtualized. In order to get a
valid performance benchmark for your workloads during your virtualization project, you
will need to develop your own workload model or benchmark. The workload model or
benchmark you develop needs to take into account your environment and its unique
requirements, characteristics, and constraints.
There are two primary methods for developing your own benchmarks to be used during
your virtualization project. Both methods are briefly discussed here. Examples of the
second method are given later in this chapter.
An easy way to get a good understanding of your workload is to pick the business day of
the month that is busiest and record all the relevant application and infrastructure
metrics so that you can reproduce the workload during your benchmark testing. You
should take seasonality into account as well, if you have a seasonal or cyclical business.
Also, make sure you adjust any numbers to meet expected cyclical peaks. If you have
this existing data available already through monitoring systems, it will make your job of
