Until now, in this article we have learned about various performance monitoring tools and commands. This recipe covers a well-known performance benchmarking tool: Sysbench. The purpose of performance benchmarking is to get a sense of system configuration and the resulting performance. Sysbench is generally used to evaluate the performance of heavy load systems. If you read the Sysbench introduction, it says that Sysbench is a benchmarking tool to evaluate a system running database under intensive load. It is also being used as a tool to evaluate the performance of multiple cloud service providers.
The current version of Sysbench supports various benchmark tests including CPU, memory, IO system, and OLTP systems. We will primarily focus on CPU, memory, and IO benchmarks.
Getting ready
Before using Sysbench, we will need to install it. Sysbench is available in the Ubuntu package repository with a little older (0.4.12) version. We will use the latest version (0.5) from Percona Systems, available in their repo.
To install Sysbench from the Percona repo, we need to add the repo to our installation sources. Following are the entries for Ubuntu 14.04 (trusty). Create a new file under /etc/apt/source.list.d and add the following lines to it:
$ sudo vi /etc/apt/sources.list.d/percona.list
deb http://repo.percona.com/apt trusty main
deb-src http://repo.percona.com/apt trusty main
Next, add the PGP key for the preceding repo:
$ sudo apt-key adv --keyserver keys.gnupg.net --recv-keys 1C4CBDCDCD2EFD2A
Now we are ready to install the latest version of Sysbench from the Percona repo. Remember to update the apt cache before installation:
$ sudo apt-get update
$ sudo apt-get install sysbench
Once installed, you can check the installed version with the --version flag to sysbench:
$ sysbench --version
sysbench 0.5
How to do it…
Now that we have Sysbench installed, let's start with performance testing our system:
Sysbench provides a prime number generation test for CPU. You can set the number of primes to be generated with the option --cpu-max-prime. Also set the limit on threads with the --num-threads option. Set the number of threads equal to the amount of CPU cores available:
$ sysbench --test=cpu --num-threads=4 \
--cpu-max-prime=20000 run
Next, we will run a test for main memory. The memory tests provides multiple options, such as block-size, total data transfer, type of memory operations, and access modes. Use the following command to run memory tests:
$ sysbench --test=memory --memory-block-size=1M \
--num-threads=2 \
--memory-total-size=100G --memory-oper=read run
Following is part of the output from the memory test:
If you have enabled huge page support, set the memory test support allocation from the huge page pool with the parameter, --memory-hugetlb. By default, it's set to off.
Next comes the storage performance test. This test also provides you with a number of options to test disk read write speeds. Depending on your requirements, you can set parameters like block-size, random or sequential read writes, synchronous or asynchronous IO operations, and many more.
For the fileio test we need a few sample files to test with. Use the sysbench prepare command to create test files. Make sure to set a total file size greater than the size of memory to avoid caching effects. I am using a small 1GBnode with 20G disk space, so I am using 15 files of 1G each:
$ sysbench --test=fileio --file-total-size=15G \
--file-num=15 prepare
Once the test preparation is complete, you can run the fileio test with different options, depending on what you want to test. The following command will perform random write operations for 60 seconds:
$ sysbench --test=fileio --file-total-size=15G \
--file-test-mode=rndwr --max-time=60 \
--file-block-size=4K --file-num=15 --num-threads=1 run
To perform random read operations, change --file-test-mode to rndrd, or to perform sequential read operations, use seqrd. You can also combine read write operations with rndrw or seqrewr. Check the help menu for more options.
When you are done with the fileio test, execute the cleanup command to delete all sample files:
$ sysbench --test=fileio cleanup
Once you have gathered various performance details, you can try updating various performance tuning parameters to boost performance. Make sure you repeat related tests after each change in parameter. Comparing results from multiple tests will help you to choose the required combination for best performance and a stable system.
There's more…
Sysbench also supports testing MySQL performance with various tests. In the same way as the fileio test, Sysbench takes care of setting a test environment by creating tables with data. When using Sysbench from the Percona repo, all OLTP test scripts are located at /usr/share/doc/sysbench/tests/db/. You will need to specify the full path when using these scripts. For example:
$ sysbench --test=oltp
The preceding command will change to the following:
$ sysbench --test=/usr/share/doc/sysbench/tests/db/ol1tp.lua
Graphing tools
Sysbench output can be hard to analyze and compare, especially with multiple runs. This is where graphs come in handy. You can try to set up your own graphing mechanism, or simply use prebuilt scripts to create graphs for you. A quick Google search gave me two good, looking options:
A Python script to extract data from Sysbench logs: https://github.com/tsuna/sysbench-tools
A shell script to extract Sysbench data to a CSV file, which can be converted to graphs: http://openlife.cc/blogs/2011/august/one-liner-condensing-sysbench-output-csv-file
More options
There are various other performance testing frameworks available. Phoronix Test Suite, Unixbench, and Perfkit by Google are some popular names. Phoronix Test Suite focuses on hardware performance and provides a wide range of performance analysis options, whereas Unixbench provides an option to test various Linux systems. Google open-sourced their performance toolkit with a benchmarker and explorer to evaluate various cloud systems.
See also
Get more details on benchmarking with Sysbench at https://wiki.mikejung.biz/Benchmarking
Sysbench documentation at http://imysql.com/wp-content/uploads/2014/10/sysbench-manual.pdf
A sample script to run batch run multiple Sysbench tests at https://gist.github.com/chetan/712484
Sysbench GitHub repo at https://github.com/akopytov/sysbench
Linux performance analysis in 60 seconds. A good read for what to check when you are debugging a performance issue at http://techblog.netflix.com/2015/11/linux-performance-analysis-in-60s.html
