sysbench安装、使用、结果解读

sysbench是一个模块化的、跨平台、多线程基准测试工具，主要用于评估测试各种不同系统参数下的数据库负载情况。
目前sysbench代码托管在github上，项目地址：https://github.com/akopytov/sysbench

一、 下载安装

git clone https://github.com/akopytov/sysbench

### RHEL/CentOS
``` shell
    yum -y install make automake libtool pkgconfig libaio-devel
    # For MySQL support, replace with mysql-devel on RHEL/CentOS 5
    yum -y install mariadb-devel openssl-devel
    # For PostgreSQL support
    yum -y install postgresql-devel

./autogen.sh
    # Add --with-pgsql to build with PostgreSQL support
    ./configure
    make -j
    make install

sysbench is a scriptable multi-threaded benchmark tool based on LuaJIT. It is most frequently used for database benchmarks, but can also be used to create arbitrarily complex workloads that do not involve a database server.

sysbench comes with the following bundled benchmarks:

• oltp_*.lua: a collection of OLTP-like database benchmarks
• fileio: a filesystem-level benchmark
• cpu: a simple CPU benchmark
• memory: a memory access benchmark
• threads: a thread-based scheduler benchmark
• mutex: a POSIX mutex benchmark

sysbench支持以下几种测试模式：

1、CPU运算性能
2、磁盘IO性能
3、调度程序性能
4、内存分配及传输速度
5、POSIX线程性能
6、数据库性能(OLTP基准测试)
目前sysbench主要支持 mysql,drizzle,pgsql,oracle 等几种数据库。

三、OLTP测试前准备
初始化测试库环境（总共10个测试表，每个表 100000 条记录，填充随机生成的数据）：

cd /tmp/sysbench-0.4.12-1.1/sysbench
mysqladmin create sbtest

./sysbench --mysql-host=1.2.3.4 --mysql-port=3317 --mysql-user=tpcc --mysql-password=tpcc \
 --test=tests/db/oltp.lua --oltp_tables_count=10 --oltp-table-size=100000 --rand-init=on prepare

关于这几个参数的解释：

--test=tests/db/oltp.lua 表示调用 tests/db/oltp.lua 脚本进行 oltp 模式测试
--oltp_tables_count=10 表示会生成 10 个测试表
--oltp-table-size=100000 表示每个测试表填充数据量为 100000
--rand-init=on 表示每个测试表都是用随机数据来填充的

如果在本机，也可以使用 –mysql-socket 指定 socket 文件来连接。加载测试数据时长视数据量而定，若过程比较久需要稍加耐心等待。

真实测试场景中，数据表建议不低于10个，单表数据量不低于500万行，当然了，要视服务器硬件配置而定。如果是配备了SSD或者PCIE SSD这种高IOPS设备的话，则建议单表数据量最少不低于1亿行。

四、进行OLTP测试

在上面初始化数据参数的基础上，再增加一些参数，即可开始进行测试了：

./sysbench --mysql-host=1.2.3.4. --mysql-port=3306 --mysql-user=tpcc \
--mysql-password=tpcc --test=tests/db/oltp.lua --oltp_tables_count=10 \
--oltp-table-size=10000000 --num-threads=8 --oltp-read-only=off \
--report-interval=10 --rand-type=uniform --max-time=3600 \
 --max-requests=0 --percentile=99 run >> ./log/sysbench_oltpX_8_20140921.log

几个选项稍微解释下

--num-threads=8 表示发起 8个并发连接
--oltp-read-only=off 表示不要进行只读测试，也就是会采用读写混合模式测试
--report-interval=10 表示每10秒输出一次测试进度报告
--rand-type=uniform 表示随机类型为固定模式，其他几个可选随机模式：uniform(固定),gaussian(高斯),special(特定的),pareto(帕累托)
--max-time=120 表示最大执行时长为 120秒
--max-requests=0 表示总请求数为 0，因为上面已经定义了总执行时长，所以总请求数可以设定为 0；也可以只设定总请求数，不设定最大执行时长
--percentile=99 表示设定采样比例，默认是 95%，即丢弃1%的长请求，在剩余的99%里取最大值

即：模拟 对10个表并发OLTP测试，每个表1000万行记录，持续压测时间为 1小时。

真实测试场景中，建议持续压测时长不小于30分钟，否则测试数据可能不具参考意义。

五、测试结果解读：

测试结果解读如下：

sysbench 0.5:  multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 8
Report intermediate results every 10 second(s)
Random number generator seed is 0 and will be ignored

Threads started!
-- 每10秒钟报告一次测试结果，tps、每秒读、每秒写、99%以上的响应时长统计
[  10s] threads: 8, tps: 1111.51, reads/s: 15568.42, writes/s: 4446.13, response time: 9.95ms (99%)
[  20s] threads: 8, tps: 1121.90, reads/s: 15709.62, writes/s: 4487.80, response time: 9.78ms (99%)
[  30s] threads: 8, tps: 1120.00, reads/s: 15679.10, writes/s: 4480.20, response time: 9.84ms (99%)
[  40s] threads: 8, tps: 1114.20, reads/s: 15599.39, writes/s: 4456.30, response time: 9.90ms (99%)
[  50s] threads: 8, tps: 1114.00, reads/s: 15593.60, writes/s: 4456.70, response time: 9.84ms (99%)
[  60s] threads: 8, tps: 1119.30, reads/s: 15671.60, writes/s: 4476.50, response time: 9.99ms (99%)
OLTP test statistics:
    queries performed:
        read:                            938224    -- 读总数
        write:                           268064    -- 写总数
        other:                           134032    -- 其他操作总数(SELECT、INSERT、UPDATE、DELETE之外的操作，例如COMMIT等)
        total:                           1340320    -- 全部总数
    transactions:                        67016  (1116.83 per sec.)    -- 总事务数(每秒事务数)
    deadlocks:                           0      (0.00 per sec.)    -- 发生死锁总数
    read/write requests:                 1206288 (20103.01 per sec.)    -- 读写总数(每秒读写次数)
    other operations:                    134032 (2233.67 per sec.)    -- 其他操作总数(每秒其他操作次数)

General statistics:    -- 一些统计结果
    total time:                          60.0053s    -- 总耗时
    total number of events:              67016    -- 共发生多少事务数
    total time taken by event execution: 479.8171s    -- 所有事务耗时相加(不考虑并行因素)
    response time:    -- 响应时长统计
         min:                                  4.27ms    -- 最小耗时
         avg:                                  7.16ms    -- 平均耗时
         max:                                 13.80ms    -- 最长耗时
         approx.  99 percentile:               9.88ms    -- 超过99%平均耗时

Threads fairness:
    events (avg/stddev):           8377.0000/44.33
    execution time (avg/stddev):   59.9771/0.00

*****  **

Build and Install

    ./autogen.sh
    # Add --with-pgsql to build with PostgreSQL support
    ./configure
    make -j
    make install

The above will build sysbench with MySQL support by default. If you have MySQL headers and libraries in non-standard locations (and no mysql_config can be found in the PATH), you can specify them explicitly with --with-mysql-includes and --with-mysql-libs options to ./configure.

To compile sysbench without MySQL support, use --without-mysql. If no database drivers are available database-related scripts will not work, but other benchmarks will be functional.

Usage

General Syntax

The general command line syntax for sysbench is:

	  sysbench [options]... [testname] [command]

• testname is an optional name of a built-in test (e.g. fileio, memory, cpu, etc.), or a name of one of the bundled Lua scripts (e.g. oltp_read_only), or a path to a custom Lua script. If no test name is specified on the command line (and thus, there is no command too, as in that case it would be parsed as a testname), or the test name is a dash ("-"), then sysbench expects a Lua script to execute on its standard input.

• command is an optional argument that will be passed by sysbench to the built-in test or script specified with testname. command defines the action that must be performed by the test. The list of available commands depends on a particular test. Some tests also implement their own custom commands.

  Below is a description of typical test commands and their purpose:

  • prepare: performs preparative actions for those tests which need them, e.g. creating the necessary files on disk for the fileio test, or filling the test database for database benchmarks.
  • run: runs the actual test specified with the testname argument. This command is provided by all tests.
  • cleanup: removes temporary data after the test run in those tests which create one.
  • help: displays usage information for the test specified with the testname argument. This includes the full list of commands provided by the test, so it should be used to get the available commands.

• options is a list of zero or more command line options starting with '--'. As with commands, the sysbench testname help command should be used to describe available options provided by a particular test.

  See General command line options for a description of general options provided by sysbench itself.

You can use sysbench --help to display the general command line syntax and options.

General Command Line Options

The table below lists the supported common options, their descriptions and default values:

Option	Description	Default value
--threads	The total number of worker threads to create	1
--events	Limit for total number of requests. 0 (the default) means no limit	0
--time	Limit for total execution time in seconds. 0 means no limit	10
--warmup-time	Execute events for this many seconds with statistics disabled before the actual benchmark run with statistics enabled. This is useful when you want to exclude the initial period of a benchmark run from statistics. In many benchmarks, the initial period is not representative because CPU/database/page and other caches need some time to warm up	0
--rate	Average transactions rate. The number specifies how many events (transactions) per seconds should be executed by all threads on average. 0 (default) means unlimited rate, i.e. events are executed as fast as possible	0
--thread-init-timeout	Wait time in seconds for worker threads to initialize	30
--thread-stack-size	Size of stack for each thread	32K
--report-interval	Periodically report intermediate statistics with a specified interval in seconds. Note that statistics produced by this option is per-interval rather than cumulative. 0 disables intermediate reports	0
--debug	Print more debug info	off
--validate	Perform validation of test results where possible	off
--help	Print help on general syntax or on a specified test, and exit	off
--verbosity	Verbosity level (0 - only critical messages, 5 - debug)	4
--percentile	sysbench measures execution times for all processed requests to display statistical information like minimal, average and maximum execution time. For most benchmarks it is also useful to know a request execution time value matching some percentile (e.g. 95% percentile means we should drop 5% of the most long requests and choose the maximal value from the remaining ones). This option allows to specify a percentile rank of query execution times to count	95
--luajit-cmd	perform a LuaJIT control command. This option is equivalent to luajit -j. See LuaJIT documentation for more information

Note that numerical values for all size options (like --thread-stack-size in this table) may be specified by appending the corresponding multiplicative suffix (K for kilobytes, M for megabytes, G for gigabytes and T for terabytes).

Random Numbers Options

sysbench provides a number of algorithms to generate random numbers that are distributed according to a given probability distribution. The table below lists options that can be used to control those algorithms.

Option	Description	Default value
--rand-type	random numbers distribution {uniform, gaussian, special, pareto, zipfian} to use by default. Benchmark scripts may choose to use either the default distribution, or specify it explictly, i.e. override the default.	special
--rand-seed	seed for random number generator. When 0, the current time is used as an RNG seed.	0
--rand-spec-iter	number of iterations for the special distribution	12
--rand-spec-pct	percentage of the entire range where 'special' values will fall in the special distribution	1
--rand-spec-res	percentage of 'special' values to use for the special distribution	75
--rand-pareto-h	shape parameter for the Pareto distribution	0.2
--rand-zipfian-exp	shape parameter (theta) for the Zipfian distribution	0.8