-<html><head><meta http-equiv="Content-Type" content="text/html; charset=ISO-8859-1"><title>dbench</title><meta name="generator" content="DocBook XSL Stylesheets V1.73.2"></head><body bgcolor="white" text="black" link="#0000FF" vlink="#840084" alink="#0000FF"><div class="refentry" lang="en"><a name="dbench.1"></a><div class="titlepage"></div><div class="refnamediv"><h2>Name</h2><p>dbench — a benchmark tool</p></div><div class="refsynopsisdiv"><h2>Synopsis</h2><div class="cmdsynopsis"><p><code class="command">dbench [OPTIONS] <num-procs></code> </p></div><div class="cmdsynopsis"><p><code class="command">dbench</code> [-B --backend=<dbench backend>] [-c --loadfile=<filename>] [-D --directory=<string>] [-F --fsync] [-R --target-rate=<double>] [-s --sync] [-S --sync-dir] [-t --timelimit=<integer>] [-T --tcp-options=<string>] [--run-once] [--fake-io] [--scsi=<scsi-device>] [--
server=<hostname>] [--export=<string>] [--protocol=<string>] [--clients-per-process=<integer>] [--trunc-io=<integer>] [--stat-check] [--skip-cleanup] [--per-client-results] [-? --help] [--usage]</p></div></div><div class="refsect1" lang="en"><a name="id2479642"></a><h2>DESCRIPTION</h2><p>
+<html><head><meta http-equiv="Content-Type" content="text/html; charset=ISO-8859-1"><title>dbench</title><meta name="generator" content="DocBook XSL Stylesheets V1.73.2"></head><body bgcolor="white" text="black" link="#0000FF" vlink="#840084" alink="#0000FF"><div class="refentry" lang="en"><a name="dbench.1"></a><div class="titlepage"></div><div class="refnamediv"><h2>Name</h2><p>dbench — a benchmark tool</p></div><div class="refsynopsisdiv"><h2>Synopsis</h2><div class="cmdsynopsis"><p><code class="command">dbench [OPTIONS] <num-procs></code> </p></div><div class="cmdsynopsis"><p><code class="command">dbench</code> [-B --backend=<dbench backend>] [-c --loadfile=<filename>] [-D --directory=<string>] [-F --fsync] [-R --target-rate=<double>] [-s --sync] [-S --sync-dir] [-t --timelimit=<integer>] [-T --tcp-options=<string>] [--run-once] [--fake-io] [--scsi=<scsi-device>] [--
allow-scsi-writes] [--server=<hostname>] [--export=<string>] [--protocol=<string>] [--clients-per-process=<integer>] [--trunc-io=<integer>] [--stat-check] [--skip-cleanup] [--per-client-results] [-? --help] [--usage]</p></div></div><div class="refsect1" lang="en"><a name="id2478376"></a><h2>DESCRIPTION</h2><p>
dbench is a utility to benchmark a system based on client workload
profiles.
- </p></div><div class="refsect1" lang="en"><a name="id247
9653"></a><h2>STANDARD OPTIONS</h2><div class="refsect2" lang="en"><a name="id2479658"></a><h3>-B --backend=<dbench backend></h3><p>
+ </p></div><div class="refsect1" lang="en"><a name="id247
8386"></a><h2>STANDARD OPTIONS</h2><div class="refsect2" lang="en"><a name="id2478392"></a><h3>-B --backend=<dbench backend></h3><p>
The backend type specifies which kind of commandset and what kind of
tests that dbench will perform. The backend type specifies which
kind of loadfile that can be used.
</p><p>
There are currently four types of backends : "fileio", "sockio", "nfs" and "scsi". The default is "fileio" which uses a CIFS style loadfile.
- </p></div><div class="refsect2" lang="en"><a name="id247
9677"></a><h3>-c --loadfile=<filename></h3><p>
+ </p></div><div class="refsect2" lang="en"><a name="id247
8410"></a><h3>-c --loadfile=<filename></h3><p>
This specifies the name of the loadfile to use. The loadfile describes
the sequence and timing of operations that dbench will issue.
- </p></div><div class="refsect2" lang="en"><a name="id247
9689"></a><h3>-D --directory=<string></h3><p>
+ </p></div><div class="refsect2" lang="en"><a name="id247
8422"></a><h3>-D --directory=<string></h3><p>
This controls which directory that dbench will use as the root for when
running the loadfile. This defaults to "." which refers to the current
directory for the "fileio" and "sockio" backends and the root of the
export for the "nfs" backend.
- </p></div><div class="refsect2" lang="en"><a name="id247
9702"></a><h3>-R --target-rate=<double></h3><p>
+ </p></div><div class="refsect2" lang="en"><a name="id247
8628"></a><h3>-R --target-rate=<double></h3><p>
By default dbench will try to replay the loadfile and keep the same
rate as the original application the loadfile was captured from.
Using this option it is possible to run the load file faster/slower
</p><p>
By setting this limit to something very large, such as 999999.99 you can
effectively tell dbench to "run this loadfile as fast as possible".
- </p></div><div class="refsect2" lang="en"><a name="id247
9732"></a><h3>-t --timelimit=<integer></h3><p>
+ </p></div><div class="refsect2" lang="en"><a name="id247
8656"></a><h3>-t --timelimit=<integer></h3><p>
How long to run the test for.
- </p></div><div class="refsect2" lang="en"><a name="id247
9741"></a><h3>--run-once</h3><p>
+ </p></div><div class="refsect2" lang="en"><a name="id247
8665"></a><h3>--run-once</h3><p>
Only run the loadfile once and stop when the end of the loadfile is
reached.
</p><p>
The default for dbench is to wrap the loadfile when the end is
reached and continue running the loadfile over and over until the
timelimit is reached.
- </p></div><div class="refsect2" lang="en"><a name="id247
9758"></a><h3>--clients-per-process=<integer></h3><p>
+ </p></div><div class="refsect2" lang="en"><a name="id247
8682"></a><h3>--clients-per-process=<integer></h3><p>
By default dbench will fork one child process for each client emulated.
Using this option dbench will run multiple emulated clients inside
each process.
This is useful for testing how performance differs between the case
of n processes with one thread of I/O each and one process with n threads
of I/O.
- </p></div><div class="refsect2" lang="en"><a name="id247
9777"></a><h3>--skip-cleanup</h3><p>
+ </p></div><div class="refsect2" lang="en"><a name="id247
8701"></a><h3>--skip-cleanup</h3><p>
Do not cleanup and delete all temporary files in the clients work
directory when the test ends.
- </p></div><div class="refsect2" lang="en"><a name="id247
9788"></a><h3>--per-client-results</h3><p>
+ </p></div><div class="refsect2" lang="en"><a name="id247
8712"></a><h3>--per-client-results</h3><p>
When the test is finished print a latency report for each inidvidual
client in addition to the aggregated report over all clients.
- </p></div></div><div class="refsect1" lang="en"><a name="id247
9800"></a><h2>FILEIO OPTIONS</h2><div class="refsect2" lang="en"><a name="id2479806"></a><h3>-F --fsync</h3><p>
+ </p></div></div><div class="refsect1" lang="en"><a name="id247
8724"></a><h2>FILEIO OPTIONS</h2><div class="refsect2" lang="en"><a name="id2478730"></a><h3>-F --fsync</h3><p>
This option only apply to the "fileio" backend.
</p><p>
This will make dbench perform a fsync() to the file after each write
operation.
- </p></div><div class="refsect2" lang="en"><a name="id247
9821"></a><h3>-s --sync-open</h3><p>
+ </p></div><div class="refsect2" lang="en"><a name="id247
8744"></a><h3>-s --sync-open</h3><p>
This option only apply to the "fileio" backend.
</p><p>
This makes dbench override the loadfile and use O_SYNC for all
file open operations.
- </p></div><div class="refsect2" lang="en"><a name="id247
9836"></a><h3>-S --sync-dir</h3><p>
+ </p></div><div class="refsect2" lang="en"><a name="id247
8760"></a><h3>-S --sync-dir</h3><p>
This option only apply to the "fileio" backend.
</p><p>
Call fsync() on the directory after each "unlink", "rmdir" or "rename"
operation. This emulates how the linux kernel nfs daemon syncs
directories after performing directory modifying operations.
- </p></div><div class="refsect2" lang="en"><a name="id2
528501"></a><h3>--fake-io</h3><p>
+ </p></div><div class="refsect2" lang="en"><a name="id2
478777"></a><h3>--fake-io</h3><p>
This option only apply to the "fileio" backend.
</p><p>
Do not do any file read/write operations at all.
- </p></div><div class="refsect2" lang="en"><a name="id2
528515"></a><h3>--stat-check</h3><p>
+ </p></div><div class="refsect2" lang="en"><a name="id2
478791"></a><h3>--stat-check</h3><p>
This option only apply to the "fileio" backend.
</p><p>
After each create/mkdir/rmdir/rename operation, immediately try to
is correct. I.e. Verify that immediately after we have created an
object that stat() will succeed and that immediately after we have
deleted an object that stat() will fail.
- </p></div></div><div class="refsect1" lang="en"><a name="id2
528535"></a><h2>SOCKIO OPTIONS</h2><div class="refsect2" lang="en"><a name="id2528541"></a><h3>-T --tcp-options=<string></h3><p>
+ </p></div></div><div class="refsect1" lang="en"><a name="id2
478812"></a><h2>SOCKIO OPTIONS</h2><div class="refsect2" lang="en"><a name="id2478817"></a><h3>-T --tcp-options=<string></h3><p>
This option only apply to the "sockio" backend.
- </p></div></div><div class="refsect1" lang="en"><a name="id2
528553"></a><h2>NFS OPTIONS</h2><div class="refsect2" lang="en"><a name="id2528558"></a><h3>--server=<hostname></h3><p>
+ </p></div></div><div class="refsect1" lang="en"><a name="id2
478829"></a><h2>NFS OPTIONS</h2><div class="refsect2" lang="en"><a name="id2478835"></a><h3>--server=<hostname></h3><p>
This option only apply to the "nfs" backend.
</p><p>
This option is mandatory when the "nfs" backend is used.
</p><p>
This specifies the host-name or ip-address of the server to test.
- </p></div><div class="refsect2" lang="en"><a name="id2
528578"></a><h3>--export=<string></h3><p>
+ </p></div><div class="refsect2" lang="en"><a name="id2
478854"></a><h3>--export=<string></h3><p>
This option only apply to the "nfs" backend.
</p><p>
This option is mandatory when the "nfs" backend is used.
</p><p>
This specifies the nfs-export on the server to do i/o to.
- </p></div><div class="refsect2" lang="en"><a name="id2
528597"></a><h3>--protocol=<string></h3><p>
+ </p></div><div class="refsect2" lang="en"><a name="id2
478873"></a><h3>--protocol=<string></h3><p>
This option only apply to the "nfs" backend.
</p><p>
This specifies whether "tcp" or "udp" is to be used. Default is "tcp".
- </p></div><div class="refsect2" lang="en"><a name="id2
528612"></a><h3>--trunc-io=<integer></h3><p>
+ </p></div><div class="refsect2" lang="en"><a name="id2
478888"></a><h3>--trunc-io=<integer></h3><p>
This option only apply to the "nfs" backend.
</p><p>
Some NFS server may have limitations on how large READ/WRITE I/Os they
accept preventing some loadfiles from running. Using this option will
override the length specified in the loadfile and make dbench never
issuing any READ/WRITE operations larger than this.
- </p></div></div><div class="refsect1" lang="en"><a name="id2528
631"></a><h2>SCSI OPTIONS</h2><div class="refsect2" lang="en"><a name="id2528637"></a><h3>--scsi=<scsi-device></h3><p>
+ </p></div></div><div class="refsect1" lang="en"><a name="id2528
498"></a><h2>SCSI OPTIONS</h2><div class="refsect2" lang="en"><a name="id2528504"></a><h3>--scsi=<scsi-device></h3><p>
This option only apply to the "scsi" backend.
</p><p>
This option is mandatory when the "scsi" backend is used.
</p><p>
This specifies the device node of the scsi-device we want to run the loadfile on. Example: --scsi=/dev/sda
- </p></div></div><div class="refsect1" lang="en"><a name="id2528659"></a><h2>COPYRIGHT/LICENSE</h2><div class="literallayout"><p><br>
+ </p></div><div class="refsect2" lang="en"><a name="id2528524"></a><h3>--allow-scsi-writes</h3><p>
+ This option only apply to the "scsi" backend.
+ </p><p>
+ By default the SCSI backend will ignore any WRITE commands in the loadfile. If you do want to performance test using real SCSI WRITE CDBs you must activate it with this argument. This WILL destroy all data on the disk!
+ </p></div></div><div class="refsect1" lang="en"><a name="id2528542"></a><h2>COPYRIGHT/LICENSE</h2><div class="literallayout"><p><br>
Copyright (C) Andrew Tridgell 2008<br>
Copyright (C) Ronnie Sahlberg 2008<br>
<br>
git.samba.org / tridge / dbench.git / commitdiff