pmstat(1) high-level system performance overview


pmstat [-gLlPxz] [-A align] [-a archive] [-h host] [-H file] [-n pmnsfile] [-O offset] [-p port] [-S starttime] [-s samples] [-T endtime] [-t interval] [-Z timezone]


pmstat provides a one line summary of system performance every interval unit of time (the default is 5 seconds). pmstat is intended to monitor system performance at the highest level, after which other tools may be used to examine subsystems in which potential performance problems may be observed in greater detail.

Multiple hosts may be monitored by supplying more than one host with multiple -h flags (for live monitoring) or by providing a name of the hostlist file, where each line contain one host name, with -H, or multiple -a flags (for retrospective monitoring from an archive).

The -t option may be used to change the default reporting interval. The interval argument follows the syntax described in PCPIntro(1), and in the simplest form may be an unsigned integer (the implied units in this case are seconds).

By default, pmstat fetches metrics by connecting to the Performance Metrics Collector Daemon (PMCD) on the local host. If the -L option is specified, then pmcd(1) is bypassed, and metrics are fetched from PMDAs on the local host using the standalone PM_CONTEXT_LOCAL variant of pmNewContext(3). When the -h option is specified, pmstat connects to the pmcd(1) on host and fetches metrics from there. As mentioned above, multiple hosts may be monitored by supplying multiple -h flags.

Alternatively, if the -a option is used, the metrics are retrieved from the Performance Co-Pilot archive log files identified by the base name archive. Multiple archives may be replayed by supplying multiple -a flags. When the -a flag is used, the -P flag may also be used to pause the output after each interval.

Standalone mode can only connect to the local host, using an archive implies a host name, and nominating a host precludes using an archive, so the options -L, -a and -h are mutually exclusive.

Normally pmstat operates on the default Performance Metrics Name Space (PMNS), however if the -n option is specified an alternative namespace is loaded from the file pmnsfile.

If the -s the option is specified, samples defines the number of samples to be retrieved and reported. If samples is 0 or -s is not specified, pmstat will sample and report continuously - this is the default behavior.

When processing an archive, pmstat may relinquish its own timing control, and operate as a ``slave'' of a pmtime(1) process that uses a GUI dialog to provide timing control. In this case, either the -g option should be used to start pmstat as the sole slave of a new pmtime(1) instance, or -p should be used to attach pmstat to an existing pmtime(1) instance via the IPC channel identified by the port argument.

The -S, -T, -O and -A options may be used to define a time window to restrict the samples retrieved, set an initial origin within the time window, or specify a ``natural'' alignment of the sample times; refer to PCPIntro(1) for a complete description of these options.

The -l option prints the last 7 characters of a hostname in summaries involving more than one host (when more than one -h option has been specified on the command line).

The -x option (extended CPU metrics) causes two additional CPU metrics to be reported, namely wait for I/O ("wa") and virtualisation steal time ("st").

The output from pmstat is directed to standard output, and the columns in the report are interpreted as follows:

The 1 minute load average.
The swpd column indicates average swap space used during the interval, in Kbytes. The free column indicates average free memory during the interval, in Kbytes. The buff column indicates average buffer memory in use during the interval, in Kbytes. The cache column indicates average cached memory in use during the interval, in Kbytes.

If the values become large, they are reported as Mbytes (m suffix) or Gbytes (g suffix).

The metrics in this area of the kernel instrumentation are of varying value. We try to report the average number of pages that are paged in (pi) and out (po) per second during the interval. If the corresponding page swapping metrics are unavailable, we report the average rate per second of swap operations in (si) and out (so) during the interval. It is normal for the ``in'' values to be non-zero, but the system is suffering memory stress if the ``out'' values are non-zero over an extended period.

If the values become large, they are reported as thousands of operations per second (K suffix) or millions of operations per second (M suffix).

The bi and bo columns indicate the average rate per second of block input and block output operations (respectfully) during the interval. Unless all file systems have a 1 Kbyte block size, these rates do not directly indicate Kbytes transfered.

If the values become large, they are reported as thousands of operations per second (K suffix) or millions of operations per second (M suffix).

Interrupt rate (in) and context switch rate (cs). Rates are expressed as average operations per second during the interval. Note that the interrupt rate is normally at least HZ (the clock interrupt rate, usually 100) interrupts per second.

If the values become large, they are reported as thousands of operations per second (K suffix) or millions of operations per second (M suffix).

Percentage of CPU time spent executing user and "nice user" code (usr), system and interrupt processing code (sys), idle loop (idl).

If any values for the associated performance metrics are unavailable, the value appears as ``?'' in the output.

By default, pmstat reports the time of day according to the local timezone on the system where pmstat is run. The -Z option changes the timezone to timezone in the format of the environment variable TZ as described in environ(5). The -z option changes the timezone to the local timezone at the host that is the source of the performance metrics, as identified via either the -h or -a options.


default PMNS specification files
pmlogger(1) configuration for creating an archive suitable for replay with pmstat


Environment variables with the prefix PCP_ are used to parameterize the file and directory names used by PCP. On each installation, the file /etc/pcp.conf contains the local values for these variables. The $PCP_CONF variable may be used to specify an alternative configuration file, as described in pcp.conf(4).


All are generated on standard error, and are intended to be self-explanatory.