How to use Performance Co-Pilot?

Solution Verified - Updated 22 Feb 2024

Environment

Red Hat Enterprise Linux (RHEL) 7 or later
Red Hat Enterprise Linux 6.6 or later (with the rhel-6-server-optional-rpms repository enabled)
Performance Co-Pilot (PCP)

Issue

How to use Performance Co-Pilot?
What are the basic steps to make Performance Co-Pilot work?
Is there a PCP crash course?

Resolution

This kbase focuses on every-day-client-commands. These are available once PCP is installed, for install details refer to Install and start using PCP.

Command Summary

The following basic commands exist within PCP:

pcp - Summarize an installation, or run a sub-command
pminfo - Display information about PCP metrics
pmval - Display the current value for named PCP metrics
pmrep - an advanced generic performance data reporting tool
pmdiff - Compare PCP archives and report significant differences
pmlogsummary - Calculate averages of metrics in PCP archives

The following commands re-implement traditional performance tools:

pcp free - Memory statistics, like free
pcp uptime - Uptime and load average statistics, like uptime
pcp numastat - Non-Uniform Memory Architecture information, like numastat
pcp vmstat - High-level system VM/CPU stats, like vmstat
pcp atop - Process information, like atop
pcp iostat - Input/output statistics for storage, like iostat
pcp pidstat - per-process status reporter (like pidstat)
pcp ipcs - IPC stats (like ipcs)

The following graphical utilities exist:

pmchart - GUI utility providing graphical display of PCP data

All PCP commands can be used in realtime on systems with PCP installed, can communicate in realtime over the network to a remote host, or can replay against pre-existing PCP log files.

Command Usage

pcp

The pcp command can be used to summarise an installation:

# pcp
Performance Co-Pilot configuration on system.example.com:

 platform: Linux system.example.com 4.2.0 #4 SMP Wed Oct 7 17:43:50 AEST 2015 x86_64
 hardware: 2 cpus, 4 disks, 1 node, 3952MB RAM
 timezone: AEST-10
 services: pmcd
     pmcd: Version 3.10.3-1, 8 agents, 1 client
     pmda: root pmcd proc xfs linux mmv netfilter jbd2
 pmlogger: primary logger: /var/log/pcp/pmlogger/system.example.com/20151009.13.15

The pcp command can also be used to run a sub-command.

The pcp free command displays memory statistics, like free:

# pcp free
             total       used       free     shared    buffers     cached
Mem:       4047228    3930064     117164          0      39600    3616772
-/+ buffers/cache:     273692    3773536
Swap       1048572        600    1047972

The pcp uptime command displays uptime and load average statistics, like uptime:

# pcp uptime
 13:24:35 up 21:39,  3 users,  load average: 0.08, 0.03, 0.05

The pcp numastat command displays Non-Uniform Memory Architecture information, like numastat:

# pcp numastat
                           node0
numa_hit               212211659
numa_miss                      0
numa_foreign                   0
interleave_hit              2575
local_node             212211659
other_node                     0

See man pcp for more information on sub-commands.

More Examples

$ pmstat -t 1 -T 3
$ pmstat -a /var/log/pcp/pmlogger/localhost.localdomain/<date-time-file-name>

$ pmatop
# recording mode
$ pmatop -w rawfile
# playback mode
$ pmatop -r rawfile

# basic system-level performance monitoring
$ pmcollectl
# subsystem specific performance monitoring
$ pmcollectl -sm
$ pmcollectl -sm –verbose
$ pmcollectl -smcdn –verbose
# record mode
$ pmcollectl -f perf-data -c 3
# playback mode:
$ pmcollectl -p perf-data

# full list of all available metrics:
$ pminfo -F
# fetch and print values for all or specific metrics:
$ pminfo -f
$ pminfo -f <metric>
$ pminfo -F -a /var/log/pcp/pmlogger/localhost.localdomain/<log-file-with-date-time>
# descriptions of all or specific metrics:
$ pminfo -T -d
$ pminfo -T -d <metric>

$ pmval <metric>
$ pmval proc.nprocs