sysbench (1)

sysbench [common-options] --test=name [test-options] command sysbench [{-h | --help} | {-v | --version}]

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.

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

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.

--num-threads

The total number of worker threads to create (defaut: 1)

--max-requests

Limit for total number of requests. 0 means unlimited (defaut: 10000)

--max-time

Limit for total execution time in seconds. 0 (defaut: 0)

--thread-stack-size

Size of stack for each thread (defaut: 32K)

--init-rnd

Specifies if random numbers generator should be initialized from timer before the test start (defaut: off)

--test

Name of the test mode to run Required

--debug

Print more debug info (default: off)

--validate

Perform validation of test results where possible (default: off)

--help

Print help on general syntax or on a test mode specified with --test, and exit

--version

Show version of program.

--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 (default: 95)

--batch

Dump current results periodically (default: off - see also the section called “Batch mode”)

--batch-delay

Delay between batch dumps in secods (default: 300 - see also the section called “Batch mode”)

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).

Batch mode

In some cases it is useful to have not only the final benchmarks statistics, but also periodical dumps of current stats to see how they change over the test run. For this purpose SysBench has a batch execution mode which is turned on by the --batch option. You may specify the delay in seconds between the consequent dumps with the --batch-delay option.

Example:

      sysbench --batch --batch-delay=5 --test=threads run

This will run SysBench in a threads test mode, with the current values of minimum, average, maximum and percentile for request execution times printed every 5 seconds.

Test modes

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

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

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:

--thread-yields

Number of lock/yield/unlock loops to execute per each request (default: 1000)

--thread-locks

Number of mutexes to create (default: 8)

Example:

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

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:

--mutex-num

Number of mutexes. The actual mutex to lock is chosen randomly before each lock (default: 4096)

--memory-scope

Possible values: global, local. Specifies whether each thread will use a globally allocated memory block, or a local one. (default: global)

--memory-total-size

Total size of data to transfer (default: 100G)

--memory-oper

Type of memory operations. Possible values: read, write

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()

Flush only data buffers, but not the metadata.

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:

--file-num

Number of files to create (default: 128)

--file-block-size

Block size to use in all I/O operations (default: 16K)

--file-total-size

Total size of files (default: 2G)

--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). (default: sync)

--file-async-backlog

Number of asynchronous operations to queue per thread (only for --file-io-mode=async, see above) (default: 128)

--file-extra-flags

Additional flags to use with open(2)

--file-fsync-freq

Do fsync() after this number of requests (default: 0 - don't use fsync())

--file-fsync-all

Do fsync() after each write operationi (default: no)

--file-fsync-end

Do fsync() at the end of the test (default: yes)

--file-fsync-mode

Which method to use for synchronization. Possible values: fsync, fdatasync (default: fsync)

--file-merged-requests

Merge at most this number of I/O requests if possible (default: 0 - don't merge)

--file-rw-ratio

reads/writes ration for combined random read/write test (default: 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.

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)

--oltp-test-mode

Execution mode (see above). Possible values: simpe (simple), complex (advanced transactional) and nontrx (non-transactional) (default: complex)

--oltp-read-only

Read-only mode. No UPDATE, DELETE or INSERT queries will be performed. (default: off)

--oltp-range-size

Range size for range queries (default: 100)

--oltp-point-selects

Number of point select queries in a single transaction (default: 10)

--oltp-simple-ranges

Number of simple range queries in a single transaction (default: 1)

--oltp-sum-ranges

Number of SUM range queries in a single transaction (default: 1)

--oltp-order-ranges

Number of ORDER range queries in a single transaction (default: 1)

--oltp-distinct-ranges

Number of DISTINCT range queries in a single transaction (default: 1)

--oltp-index-updates

Number of index UPDATE queries in a single transaction (default: 1)

--oltp-non-index-updates

Number of non-index UPDATE queries in a single transaction (default: 1)

--oltp-nontrx-mode

Type of queries for non-transactional execution mode (see above). Possible values: select, update_key, update_nokey, insert, delete. (default: select)

--oltp-connect-delay

Time in microseconds to sleep after each connection to database (default: 10000)

--oltp-user-delay-min

Minimum time in microseconds to sleep after each request (default: 0)

--oltp-user-delay-max

Maximum time in microseconds to sleep after each request (default: 0)

--oltp-table-name

Name of the test table (default: sbtest)

--oltp-table-size

Number of rows in the test table (default: 10000)

--oltp-dist-type

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

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

--oltp-dist-pct

Percentage of values to be treated as 'special' (for special distribution) (default: 1)

--oltp-dist-res

Percentage of cases when 'special' values are generated (for special distribution) (default: 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. (default: auto)

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

--mysql-host

MySQL server host. (default: localhost)

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.

--mysql-port

MySQL server port (in case TCP/IP connection should be used) (default: 3306)

--mysql-socket

Unix socket file to communicate with the MySQL server

--mysql-user

MySQL user (default: 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. (default: sbtest)

--mysql-table-engine

Type of the test table. Possible values: myisam, innodb, heap, ndbcluster, bdb, maria, falcon, pbxt (default: innodb)

--mysql-ssl

Use SSL connections. (default: no)

--myisam-max-rows

MAX_ROWS option for MyISAM tables (required for big tables) (default: 1000000)

--mysql-create-options

Additional options passed to CREATE TABLE.

Example usage:

	      $ sysbench --test=oltp --mysql-table-type=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 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.

Alexey Kopytov <[email protected]>

Author.

Copyright © 2004-2008 MySQL AB

This manual page was rewritten for the Debian system (and may be used by others) from the manual.xml of the original package.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU General Public License, Version 2 or (at your option) any later version published by the Free Software Foundation.

On Debian systems, the complete text of the GNU General Public License can be found in /usr/share/common-licenses/GPL.