Supporting Hardware - Expert Consolidation in Oracle Database 12c - page 90

written, can represent your workload. If your applications perform well on the new platform, then there is a higher
probability of the platform's acceptance and ultimate success. The worst possible outcome for the consolidated
platform would be the lack of acceptance!

■ Some of the benchmarks listed and described below can perform destructive testing and/or cause high
system load. always ensure that your benchmark does not have any negative impact on your environment! do not
benchmark outside your controlled and isolated lab environment! always carefully read the documentation that comes
with the benchmark. and be especially careful when it comes to writing benchmarks

Caution

If you cannot find representative workloads from within the range of applications in use in your organization,
you may have to resort to an off-the-shelf benchmark. Large Enterprise Resource Planning systems usually come with
a benchmark, but those systems are unlikely candidates for your consolidation platform. If you want to test different
components of your system, specialized benchmarks come to mind, but those don't test the platform end-to-end.
For storage related testing you could use these benchmarks, for example:

• iozone

• bonnie++

• hdbench

• fio

• And countless others more . . .

Of all the available storage benchmarks, FIO sounds very interesting. It is a very flexible benchmark written by
Jens Axboe to test different I/O schedulers in Linux and will be presented in more detail below.
Network-related benchmarks, for example, include the officially Oracle-sanctioned iperf and others. Kevin
Closson's Silly Little Benchmark tests memory performance, and so does the University of Virginia's STREAM
benchmark. The chips on the mainboard, including the CPU and also cooling, can be tested using the High
Performance Linpack benchmark, which is often used in High Performance Computing (HPC).
Each of these benchmarks can be used to get the bigger picture, but none is really suited to assess the system's
qualities when used in isolation. More information about the suitability of the platform, especially in respect to
storage performance, can be obtained by using Oracle IO Numbers (ORION) or the Silly Little Oracle Benchmark
(SLOB) written by Kevin Closson. The below sections are a small selection of available benchmarks.

FIO
FIO is an intelligent I/O testing platform written by Jens Axboe, whom we also have to thank for the Linux I/O
schedulers and much more. Out of all the I/O benchmarks, I like FIO because it is very flexible and offers a wealth of
output for the performance analyst. As with most of I/O related benchmarks, FIO works best in conjunction with other
tools to get a more complete picture.
What is great about the tool is the flexibility, but it requires a little more time to learn all the ins and outs of it.
However, if you take the time to learn FIO, you will automatically learn more about Linux as well. FIO benchmark runs
are controlled by plain text files with instructions, called a job file. A sample file is shown here; it will be used later in
the chapter:

[oracle@server1 fio-2.1]# cat rand-read-sync.fio
[random-read-sync]
rw=randread