mirror of
https://github.com/10h30/lookbusy.git
synced 2026-05-12 15:21:20 +09:00
flow2000
72 lines
3.8 KiB
Markdown
72 lines
3.8 KiB
Markdown
## lookbusy 1.4 -- a synthetic load generator for Linux systems
|
|
|
|
This is lookbusy, a tool for making systems busy. It uses relatively simple
|
|
techniques to generate CPU activity, memory and disk utilization and traffic.
|
|
|
|
lookbusy is not a benchmarking tool, or a realistic load simulator. While it
|
|
attempts to produce load factors which are exhibited by real applications, the
|
|
exact operations used are not modelled on real applications, and at the low
|
|
level, the exact hardware operations are not identical.
|
|
|
|
* Process Structure
|
|
|
|
One lookbusy process is forked for each load-generation task -- that is, one
|
|
process per CPU, one for memory usage, and one for each file on disk being
|
|
used, plus a toplevel parent process. Errors in or termination of any process
|
|
will trigger a shutdown in all others. It's safe to use ^C from a terminal,
|
|
or to kill processes remotely.
|
|
|
|
* CPU Usage Modes
|
|
|
|
Two CPU usage modes are provided. The first, 'fixed', attempts to keep total
|
|
CPU utilization at a particular level, using up any balance in idle time
|
|
between the other processes on the host and the preferred level (if other
|
|
processes are themselves able to exceed the chosen level, obviously, lookbusy
|
|
can't fix that situation but will drop its own usage to near zero and wait for
|
|
load to drop.)
|
|
|
|
The second mode, 'curve', produces utilization levels which vary over a chosen
|
|
range, over a given interval. The simplest (and the default) usage is to
|
|
modulate usage smoothly over the course of a 24-hour period, peaking at local
|
|
midnight and bottoming out at local noon. Options are provided to adjust all
|
|
these settings -- see lookbusy(1).
|
|
|
|
* CPU Concurrency
|
|
|
|
lookbusy has basic awareness of multiprocessor and multi-logical-CPU systems;
|
|
it will attempt to keep cumulative system usage at the chosen level by forking
|
|
multiple instances of itself, one per CPU. CPUs with a nonzero physical-id,
|
|
such as are found on hyperthreaded i386 CPUs, are ignored when counting. The
|
|
CPU utilization algorithm uses a tight arithmetic loop, which should be
|
|
entirely register-based on most CPUs, incurring no memory traffic.
|
|
|
|
* Portability
|
|
|
|
As of 1.0, lookbusy claims support only for Linux systems. Most of its
|
|
implementation is entirely portable to other UNIX systems; memory and disk
|
|
usage should work as-is. CPU utilization will almost certainly need porting
|
|
work, as concerns use of the /proc filesystem and handling of SMP.
|
|
|
|
|
|
|
|
$Id$
|
|
|
|
## General options
|
|
|
|
-h, --help Commandline help (you're reading it)
|
|
-v, --verbose Verbose output (may be repeated)
|
|
-q, --quiet Be quiet, produce output on errors only
|
|
CPU usage options:
|
|
-c, --cpu-util=PCT, Desired utilization of each CPU, in percent (default --cpu-util=RANGE 50%). If 'curve' CPU usage mode is chosen, a range of the form MIN-MAX should be given.
|
|
-n, --ncpus=NUM Number of CPUs to keep busy (default: autodetected)
|
|
-r, --cpu-mode=MODE Utilization mode ('fixed' or 'curve', see lookbusy(1))
|
|
-p, --cpu-curve-peak=TIME Offset of peak utilization within curve period, in seconds (append 'm', 'h', 'd' for other units)
|
|
-P, --cpu-curve-period=TIME Duration of utilization curve period, in seconds (append 'm', 'h', 'd' for other units)
|
|
Memory usage options:
|
|
-m, --mem-util=SIZE Amount of memory to use (in bytes, followed by KB, MB, or GB for other units; see lookbusy(1))
|
|
-M, --mem-sleep=TIME Time to sleep between iterations, in usec (default 1000)
|
|
Disk usage options:
|
|
-d, --disk-util=SIZE Size of files to use for disk churn (in bytes, followed by KB, MB, GB or TB for other units)
|
|
-b, --disk-block-size=SIZE Size of blocks to use for I/O (in bytes, followed by KB, MB or GB)
|
|
-D, --disk-sleep=TIME Time to sleep between iterations, in msec (default 100)
|
|
-f, --disk-path=PATH Path to a file/directory to use as a buffer (default /tmp); specify multiple times for additional paths |