1hackbench(8) hackbench(8)
2
6 hackbench - scheduler benchmark/stress test
7
9 hackbench [-p|--pipe] [-s|--datasize <bytes>] [-l|--loops <num-loops>]
10 [-g|--groups <num-groups>] [-f|--fds <num-fds>] [-T|--threads]
11 [-P|--process] [-F|--fifo] [--help]
12
15 Hackbench is both a benchmark and a stress test for the Linux kernel
16 scheduler. It's main job is to create a specified number of pairs of
17 schedulable entities (either threads or traditional processes) which
18 communicate via either sockets or pipes and time how long it takes for
19 each pair to send data back and forth.
20
23 These programs follow the usual GNU command line syntax, with long
24 options starting with two dashes ("--").
25 A summary of options is included below.
26 -p, --pipe
28 Sends the data via a pipe instead of the socket (default)
29 -s, --datasize=<size in bytes>
31 Sets the amount of data to send in each message
32 -l, --loops=<number of loops>
34 How many messages each sender/receiver pair should send
35 -g, --groups=<number of groups>
37 Defines how many groups of senders and receivers should be
38 started
39 -f, --fds=<number of file descriptors>
41 Defines how many file descriptors each child should use. Note
42 that the effective number will be twice the amount you set here,
43 as the sender and receiver children will each open the given
44 amount of file descriptors.
45 -T, --threads
47 Each sender/receiver child will be a POSIX thread of the parent.
48 -P, --process
50 Hackbench will use fork() on all children (default behaviour)
51 -F,--fifo
53 Change the main thread to SCHED_FIFO after creating workers.
54 This allows the management thread to run after many workers are
55 created.
56 --help
58 Shows a simple help screen
59
61 Running hackbench without any options will give default behaviour,
62 using fork() and sending data between senders and receivers via sock‐
63 ets.
64 user@host: ~ $ hackbench
66 Running in process mode with 10 groups using 40 file descriptors each
67 (== 400 tasks)
68 Each sender will pass 100 messages of 100 bytes
69 Time: 0.890
70 To use pipes between senders and receivers and using threads instead of
72 fork(), run
73 user@host: ~ $ hackbench --pipe --threads (or hackbench -p -T)
75 Running in threaded mode with 10 groups using 40 file descriptors each
76 (== 400 tasks)
77 Each sender will pass 100 messages of 100 bytes
78 Time: 0.497
79 Set the datasize to 512 bytes, do 200 messages per sender/receiver
81 pairs and use 15 groups using 25 file descriptors per child, in process
82 mode.
83 user@host: ~ $ hackbench -s 512 -l 200 -g 15 -f 25 -P
85 Running in process mode with 15 groups using 50 file descriptors each
86 (== 750 tasks)
87 Each sender will pass 200 messages of 512 bytes
88 Time: 4.497
89
91 hackbench was written by Rusty Russell <rusty@rustcorp.com.au> with
92 contributions from Yanmin Zhang <yanmin_zhang@linux.intel.com>, Ingo
93 Molnar <mingo@elte.hu> and David Sommerseth <davids@redhat.com>
94 This manual page was written by Clark Williams <williams@redhat.com>
96 and David Sommerseth <davids@redhat.com>
97
99 This version of hackbench is based on the code downloaded from
100 http://people.redhat.com/mingo/cfs-scheduler/tools/hackbench.c. Yanmin
101 Zhang merged the original hackbench code from
102 http://devresources.linuxfoundation.org/craiger/hackbench/src/hack‐
103 bench.c which uses fork() and a modified version from
104 http://www.bullopensource.org/posix/pi-futex/hackbench_pth.c which uses
105 pthread only and gave the possibility to change behaviour at run time.
106 Hackbench have since then gone through some more rewriting to improve
107 error handling and proper tracking of fork()ed children, to avoid leav‐
108 ing zombies on the system if hackbench stops unexpectedly.
109 February 23, 2010 hackbench(8)