SysBench manual

Alexey Kopytov


Table of Contents

1. About this document
1. Translations
2. Acknowledgments
2. Introduction
1. Features of SysBench
2. Design
3. Links
4. Installation
3. Usage
1. General syntax
2. General command line options
3. Test modes
3.1. cpu
3.2. threads
3.3. mutex
3.4. memory
3.5. fileio
3.6. oltp

Chapter 1. About this document

This document is a user manual for SysBench, a multi-threaded benchmark tool available from http://launchpad.net/sysbench. This document describes features provided by the 0.4.x development branch of SysBench. New features available in the newer 0.5.x branch are not covered by this document.

1.  Translations

The following translations of this document are currently available:

2.  Acknowledgments

Thanks to Vera Djuraskovic for contributing a Serbo-Croatian translation of this document.

Chapter 2. Introduction

SysBench is a modular, cross-platform and multi-threaded benchmark tool for evaluating OS parameters that are important for a system running a database under intensive load.

The idea of this benchmark suite is to quickly get an impression about system performance without setting up complex database benchmarks or even without installing a database at all.

1. Features of SysBench

Current features allow to test the following system parameters:

  • file I/O performance

  • scheduler performance

  • memory allocation and transfer speed

  • POSIX threads implementation performance

  • database server performance

2. Design

The design is very simple. SysBench runs a specified number of threads and they all execute requests in parallel. The actual workload produced by requests depends on the specified test mode. You can limit either the total number of requests or the total time for the benchmark, or both.

Available test modes are implemented by compiled-in modules, and SysBench was designed to make adding new test modes an easy task. Each test mode may have additional (or workload-specific) options.

4. Installation

If you are building SysBench from a Bazaar repository rather than from a release tarball, you should run ./autogen.sh before building.

The following standart procedure will be sufficient to build SysBench in most cases:

	  ./configure
	  make
	  make install
	

The above procedure will try to compile 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 environmental variable), then 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. In this case all database-related test modes will be unavailable.

If you are running on a 64-bit platform, make sure to build a 64-bit binary by passing the proper target platform and compiler options to configure script. You can also consult the INSTALL file for generic installation instructions.

Chapter 3. Usage

1. General syntax

The general syntax for SysBench is as follows:

	  sysbench [common-options] --test=name [test-options] command
	

See Section 2, “General command line options” for a description of common options and documentation for particular test mode for a list of test-specific options.

Below is a brief description of available 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 the oltp test.
run
Runs the actual test specified with the --test=name option.
cleanup
Removes temporary data after the test run in those tests which create one.
help
Displays usage information for a test specified with the --test=name option.

Also you can use sysbench help to display the brief usage summary and the list of available test modes.

2. General command line options

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

Option Description Default value
--num-threads The total number of worker threads to create 1
--max-requests Limit for total number of requests. 0 means unlimited 10000
--max-time Limit for total execution time in seconds. 0 (default) means unlimited 0
--forced-shutdown

Amount of time to wait after --max-time before forcing shutdown. The value can be either an absolute number of seconds or as a percentage of the --max-time value by specifying a number of percents followed by the '%' sign.

"off" (the default value) means that no forced shutdown will be performed.

off
--thread-stack-size Size of stack for each thread 32K
--init-rng Specifies if random numbers generator should be initialized from timer before the test start off
--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
--test Name of the test mode to run Required
--debug Print more debug info off
--validate Perform validation of test results where possible off
--help Print help on general syntax or on a test mode specified with --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
--validate Perform validation of test results where possible off

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).

3. Test modes

This section gives a detailed description for each test mode available in SysBench.

3.1. cpu

The cpu is one of the most simple benchmarks in SysBench. In this mode each request consists in calculation of prime numbers up to a value specified by the --cpu-max-primes option. All calculations are performed using 64-bit integers.

Each thread executes the requests concurrently until either the total number of requests or the total execution time exceed the limits specified with the common command line options.

Example:

	  sysbench --test=cpu --cpu-max-prime=20000 run
	

3.2. threads

This test mode was written to benchmark scheduler performance, more specifically the cases when a scheduler has a large number of threads competing for some set of mutexes.

SysBench creates a specified number of threads and a specified number of mutexes. Then each thread starts running the requests consisting of locking the mutex, yielding the CPU, so the thread is placed in the run queue by the scheduler, then unlocking the mutex when the thread is rescheduled back to execution. For each request, the above actions are run several times in a loop, so the more iterations is performed, the more concurrency is placed on each mutex.

The following options are available in this test mode:

OptionDescriptionDefault value
--thread-yieldsNumber of lock/yield/unlock loops to execute per each request1000
--thread-locksNumber of mutexes to create8

Example:

	  sysbench --num-threads=64 --test=threads --thread-yields=100 --thread-locks=2 run
	

3.3. mutex

This test mode was written to emulate a situation when all threads run concurrently most of the time, acquiring the mutex lock only for a short period of time (incrementing a global variable). So the purpose of this benchmarks is to examine the performance of mutex implementation.

The following options are available in this test mode:

OptionDescriptionDefault value
--mutex-numNumber of mutexes. The actual mutex to lock is chosen randomly before each lock4096
--mutex-locksNumber of mutex locks to acquire per each request50000
--mutex-loopsNumber of iterations for an empty loop to perform before acquiring the lock10000

3.4. memory

This test mode can be used to benchmark sequential memory reads or writes. Depending on command line options each thread can access either a global or a local block for all memory operations.

The following options are available in this test mode:

OptionDescriptionDefault value
--memory-block-sizeSize of memory block to use1K
--memory-scope Possible values: global, local. Specifies whether each thread will use a globally allocated memory block, or a local one. global
--memory-total-sizeTotal size of data to transfer100G
--memory-oper Type of memory operations. Possible values: read, write. 100G

3.5. fileio

This test mode can be used to produce various kinds of file I/O workloads. At the prepare stage SysBench creates a specified number of files with a specified total size, then at the run stage, each thread performs specified I/O operations on this set of files.

When the global --validate option is used with the fileio test mode, SysBench performs checksums validation on all data read from the disk. On each write operation the block is filled with random values, then the checksum is calculated and stored in the block along with the offset of this block within a file. On each read operation the block is validated by comparing the stored offset with the real offset, and the stored checksum with the real calculated checksum.

The following I/O operations are supported:

seqwr
sequential write
seqrewr
sequential rewrite
seqrd
sequential read
rndrd
random read
rndwr
random write
rndrw
combined random read/write

Also, the following file access modes can be specified, if the underlying platform supports them:

Asynchronous I/O mode
At the moment only Linux AIO implementation is supported. When running in asynchronous mode, SysBench queues a specified number of I/O requests using Linux AIO API, then waits for at least one of submitted requests to complete. After that a new series of I/O requests is submitted.
Slow mmap() mode
In this mode SysBench will use mmap'ed I/O. However, a separate mmap will be used for each I/O request due to the limitation of 32-bit architectures (we cannot mmap() the whole file, as its size migth possibly exceed the maximum of 2 GB of the process address space).
Fast mmap() mode
On 64-bit architectures it is possible to mmap() the whole file into the process address space, avoiding the limitation of 2 GB on 32-bit platforms.
Using fdatasync() instead of fsync()
Additional flags to open(2)
SysBench can use additional flags to open(2), such as O_SYNC, O_DSYNC and O_DIRECT.

Below is a list of test-specific option for the fileio mode:

OptionDescriptionDefault value
--file-numNumber of files to create128
--file-block-size Block size to use in all I/O operations 16K
--file-total-sizeTotal size of files2G
--file-test-mode Type of workload to produce. Possible values: seqwr, seqrewr, seqrd, rndrd, rndwr, rndwr (see above) required
--file-io-mode I/O mode. Possible values: sync, async, fastmmap, slowmmap (only if supported by the platform, see above). sync
--file-async-backlog Number of asynchronous operations to queue per thread (only for --file-io-mode=async, see above) 128
--file-extra-flags Additional flags to use with open(2)  
--file-fsync-freq Do fsync() after this number of requests (0 - don't use fsync()) 100
--file-fsync-all Do fsync() after each write operation no
--file-fsync-end Do fsync() at the end of the test yes
--file-fsync-mode Which method to use for synchronization. Possible values: fsync, fdatasync (see above) fsync
--file-merged-requests Merge at most this number of I/O requests if possible (0 - don't merge) 0
--file-rw-ratio reads/writes ration for combined random read/write test 1.5

Usage example:

	    $ sysbench --num-threads=16 --test=fileio --file-total-size=3G --file-test-mode=rndrw prepare
	    $ sysbench --num-threads=16 --test=fileio --file-total-size=3G --file-test-mode=rndrw run
	    $ sysbench --num-threads=16 --test=fileio --file-total-size=3G --file-test-mode=rndrw cleanup
	  

In the above example the first command creates 128 files with the total size of 3 GB in the current directory, the second command runs the actual benchmark and displays the results upon completion, and the third one removes the files used for the test.

3.6. oltp

This test mode was written to benchmark a real database performance. At the prepare stage the following table is created in the specified database (sbtest by default):

	  CREATE TABLE `sbtest` (
	  `id` int(10) unsigned NOT NULL auto_increment,
	  `k` int(10) unsigned NOT NULL default '0',
	  `c` char(120) NOT NULL default '',
	  `pad` char(60) NOT NULL default '',
	  PRIMARY KEY  (`id`),
	  KEY `k` (`k`);
	

Then this table is filled with a specified number of rows.

The following execution modes are available at the run stage:

Simple

In this mode each thread runs simple queries of the following form:

 SELECT c FROM sbtest WHERE id=N  

where N takes a random value in range 1..<table size>

Advanced transactional

Each thread performs transactions on the test table. If the test table and database support transactions (e.g. InnoDB engine in MySQL), then BEGIN/COMMIT statements will be used to start/stop a transaction. Otherwise, SysBench will use LOCK TABLES/UNLOCK TABLES statements (e.g. for MyISAM engine in MySQL). If some rows are deleted in a transaction, the same rows will be inserted within the same transaction, so this test mode does not destruct any data in the test table and can be run multiple times on the same table.

Depending on the command line options, each transaction may contain the following statements:
  • Point queries:
    SELECT c FROM sbtest WHERE id=N
  • Range queries:
    SELECT c FROM sbtest WHERE id BETWEEN N AND M 
  • Range SUM() queries:
    SELECT SUM(K) FROM sbtest WHERE id BETWEEN N and M
  • Range ORDER BY queries:
    SELECT c FROM sbtest WHERE id between N and M ORDER BY c
  • Range DISTINCT queries:
    SELECT DISTINCT c FROM sbtest WHERE id BETWEEN N and M ORDER BY c
  • UPDATEs on index column:
    UPDATE sbtest SET k=k+1 WHERE id=N 
  • UPDATEs on non-index column:
    UPDATE sbtest SET c=N WHERE id=M 
  • DELETE queries:
    DELETE FROM sbtest WHERE id=N 
  • INSERT queries:
    INSERT INTO sbtest VALUES (...) 
Non-transactional

This mode is similar to Simple, but you can also choose the query to run. Note that unlike the Advanced transactional mode, this one does not preserve the test table between requests, so you should recreate it with the appropriate cleanup/prepare commands between consecutive benchmarks.

Below is a list of possible queries:

  • Point queries:
    SELECT pad FROM sbtest WHERE id=N
  • UPDATEs on index column:
    UPDATE sbtest SET k=k+1 WHERE id=N
  • UPDATEs on non-index column:
    UPDATE sbtest SET c=N WHERE id=M
  • DELETE queries:
    DELETE FROM sbtest WHERE id=N
    The generated row IDs are unique over each test run, so no row is deleted twice.
  • INSERT queries:
    INSERT INTO sbtest (k, c, pad) VALUES(N, M, S)

Below is a list of options available for the database test mode:

OptionDescriptionDefault value
--oltp-test-modeExecution mode (see above). Possible values: simpe (simple), complex (advanced transactional) and nontrx (non-transactional)complex
--oltp-read-only Read-only mode. No UPDATE, DELETE or INSERT queries will be performed. off
--oltp-skip-trx Omit BEGIN/COMMIT statements, i.e. run the same queries as the test would normally run but without using transactions. off
--oltp-reconnect-mode Reconnect mode. Possible values:
sessionDon't reconnect (i.e. each thread disconnects only at the end of the test)
queryReconnect after each SQL query
transactionReconnect after each transaction (if transactions are used in the selected DB test)
randomOne of the above modes is randomly chosen for each transaction
session
--oltp-range-sizeRange size for range queries100
--oltp-point-selects Number of point select queries in a single transaction 10
--oltp-simple-ranges Number of simple range queries in a single transaction 1
--oltp-sum-ranges Number of SUM range queries in a single transaction 1
--oltp-order-ranges Number of ORDER range queries in a single transaction 1
--oltp-distinct-ranges Number of DISTINCT range queries in a single transaction 1
--oltp-index-updates Number of index UPDATE queries in a single transaction 1
--oltp-non-index-updates Number of non-index UPDATE queries in a single transaction 1
--oltp-nontrx-mode Type of queries for non-transactional execution mode (see above). Possible values: select, update_key, update_nokey, insert, delete. select
--oltp-connect-delay Time in microseconds to sleep after each connection to database 10000
--oltp-user-delay-min Minimum time in microseconds to sleep after each request 0
--oltp-user-delay-max Maximum time in microseconds to sleep after each request 0
--oltp-table-name Name of the test table sbtest
--oltp-table-size Number of rows in the test table 10000
--oltp-dist-type

Distribution of random numbers. Possible values: uniform (uniform distribution), gauss (gaussian distribution) and special.

With special distribution a specified percent of numbers is generated in a specified percent of cases (see options below).

special
--oltp-dist-pct Percentage of values to be treated as 'special' (for special distribution) 1
--oltp-dist-res Percentage of cases when 'special' values are generated (for special distribution) 75
--db-ps-mode If the database driver supports Prepared Statements API, SysBench will use server-side prepared statements for all queries where possible. Otherwise, client-side (or emulated) prepared statements will be used. This option allows to force using emulation even when PS API is available. Possible values: disable, auto. auto

Also, each database driver may provide its own options. Currently only MySQL driver is available. Below is a list of MySQL-specific options:

OptionDescriptionDefault value
--mysql-host

MySQL server host.

Starting from version 0.4.5 you may specify a list of hosts separated by commas. In this case SysBench will distribute connections between specified MySQL hosts on a round-robin basis. Note that all connection ports and passwords must be the same on all hosts. Also, databases and tables must be prepared explicitely on each host before executing the benchmark.

localhost
--mysql-port MySQL server port (in case TCP/IP connection should be used) 3306
--mysql-socketUnix socket file to communicate with the MySQL server 
--mysql-user MySQL user user
--mysql-password MySQL password  
--mysql-db MySQL database name. Note SysBench will not automatically create this database. You should create it manually and grant the appropriate privileges to a user which will be used to access the test table. sbtest
--mysql-table-engine Type of the test table. Possible values: myisam, innodb, heap, ndbcluster, bdb, maria, falcon, pbxt innodb
--mysql-ssl Use SSL connections. no
--myisam-max-rows MAX_ROWS option for MyISAM tables (required for big tables) 1000000
--mysql-create-options Additional options passed to CREATE TABLE.  

Example usage:

	  $ sysbench --test=oltp --mysql-table-engine=myisam --oltp-table-size=1000000 --mysql-socket=/tmp/mysql.sock prepare
	  $ sysbench --num-threads=16 --max-requests=100000 --test=oltp --oltp-table-size=1000000 --mysql-socket=/tmp/mysql.sock --oltp-read-only=on run
	

The first command will create a MyISAM table 'sbtest' in a database 'sbtest' on a MySQL server using /tmp/mysql.sock socket, then fill this table with 1M records. The second command will run the actual benchmark with 16 client threads, limiting the total number of request by 100,000.