Go to main content

man pages section 8: System Administration Commands

Exit Print View

Updated: Thursday, June 13, 2019
 
 

pgstat(8)

Name

pgstat - report utilization statistics for Processor Groups

Synopsis

pgstat [-A] [-B core|soc|sys] [-C] [-l] [-p] [-s key | -S key]
    [-t number] [-T u | d] [-v] [-r string] [-R string]
     [-P pg ...] [-c processor_id... ] [interval [count]]
pgstat -h

Description

The pgstat displays utilization statistics about Processor Groups (PGs). A PG is a set of CPUs that are grouped together by a common characteristic.

PGs are used by the operating system to represent CPUs that share performance relevant hardware, such as execution pipelines, caches, and so forth. These PGs are organized into a hierarchy that models the processor topology of the machine. In this hierarchy, each CPU (strand) has a leaf PG that represents the CPUs that share the most hardware with it. Each successive ancestor of the leaf PG shares progressively less hardware with the CPU until the root PG is reached. The root PG contains all of the CPUs in the system and represents the group of CPUs sharing the least hardware with each other. (See “Examples” below for an example of PG hierarchy).

If a machine does not have any performance-relevant hardware sharing relationships, then pgstat displays only a root PG that contains all of the CPUs in the system.

By default, pgstat does the following:

  • Measures the hardware and software utilization of all PGs in the PG hierarchy over a one second interval.

  • Displays the utilization of the PGs in depth first order using indentation to help show how the PGs relate to each other.

  • Displays the ID, sharing relationship, hardware load, software load, and online CPUs for each PG at the end of each interval.

The interval and count can be given as arguments to specify the number of seconds in the sampling interval and number of times to measure and display the utilization for the specified PGs.

You can specify options to further tailor the output, organize the output a certain way, and specify PGs of interest (see “Options” below for details).

A hyphen (“-”) is displayed when the utilization for a given PG is not supported and a question mark (?) is displayed when the utilization is not available. On systems where the CPU hardware performance counters are needed to measure the hardware utilization, the hardware utilization might be unavailable because the counters are being used by a cpc(3CPC) consumer such as attributes(7), cputrack(1), dtrace(8), or another application that uses libcpc(3LIB).

Options

The following options are supported:

–A

Display summary of utilization data when pgstat is run over multiple intervals.

–B core

Bin and display utilization data by core ID. This option is incompatible with the –P option.

–B soc

Bin and display utilization data by socket ID. This option is incompatible with the –P option.

–B sys

Bin and display utilization for the entire system. This option is incompatible with the –P option.

–c processor_id...

Display utilization about PGs that contain the specified CPUs. The CPUs can be specified as a comma separated list of CPU IDs. A hyphen (“-”) can be used to specify contiguous ranges of CPU IDs (for example, 0-3).

–C

Display utilization of each CPU in each PG.

–h

Display short help message and exit with exit status 0.

–l

Use the learned capacity in hardware utilization calculations if it is larger than the platform's advertised capacity. The learned capacity is the maximum observed hardware utilization since starting pgstat, and is useful in situations where the platform does not provide an accurate maximum capacity.

–p

Display the physical relationship that corresponds to a PG. If a PG has the same CPUs as the whole system, a processor core, or a socket, system, core, or socket, as appropriate along with the corresponding ID, is displayed after the sharing relationship of the PG in square brackets (“[]”).

–P pg,...

Display utilization for specified PGs. Multiple PGs can be specified as a comma-separated list of PG IDs. A hyphen (“-”) can be used to specify a contiguous range of PG IDs (for example, 0-3).

–r string1,string2,...

Display utilization only for PGs with a sharing relationship name that matches any of the specified strings. The string can be a whole relationship name or a portion of one or more relationship names. The string matching is case-insensitive.

Multiple –r options can be entered, which results in matching all PGs with a relationship name that matches any of the specified strings.

–R string1,string2,...

Display information only about PGs with a sharing relationship name other than the one(s) specified.

String matching is the same as described above for the –r option. Multiple –R options can be entered.

–s key

Sort output lines by the specified key in descending order. The specified key can be one of the following:

pg

Sort by PG ID.

hwload

Sort by hardware utilization.

swload

Sort by software utilization.

user

Sort by user time.

sys

Sort by system time.

idle

Sort by idle time.

depth

Sort by descending PG tree from root to leaves, depth-first (default).

breadth

Sort by descending PG tree from root to leaves, breadth-first.

–S key

Sort output lines by the specified key in ascending order. Possible key values are the same as for the –s option.

–t number

Show the top number of PGs for the specified integer number.

–T u | d

Display timestamp for each sampling interval in Unix time (see time(2)) or the standard date format used by date(1).

–v

Display extra information about each PG including hardware utilization and capacity and software user, system, and idle times.

Output Headings

The pgstat command displays the column headings, which are listed below, along with the meanings of those headings.

PG

Processor Group ID.

RELATIONSHIP

Sharing relationship for PG.

HW

Hardware load in percent (calculated as UTIL/ CAP for interval).

UTIL

Hardware utilization of PG's shared hardware component normalized per second. This can be a large number, so K, M, B, and T are used for denoting thousand, million, billion, and trillion, respectively.

CAP

Approximate maximum possible utilization for PG's shared hardware component per second. This value is fixed by default at a best guess for the given platform, but can be changed to a learned value using the –l option. This can be a large number, so K, M, B, and T are used for denoting thousand, million, billion, and trillion, respectively.

SW

Software load in percent (calculated as (USR + SYS) / (USR + SYS + IDLE ))

USR

Percentage of time that software threads ran in user mode on CPUs in PG during interval.

SYS

Percentage of time that software threads ran in system mode on CPUs in PG during interval.

IDLE

Percentage of time that no software threads ran on CPUs in PG during interval.

CPUS

CPU IDs for CPUs in PG.

Examples

In the following examples, the system contains one UltraSPARC T1 processor chip with 8 cores and 32 strands.

Example 1 Displaying Utilization for Specified Period

The following command displays utilization for all PGs over the last two seconds.

$ pgstat 1 2
PG  RELATIONSHIP            HW    SW  CPUS
 0  System                   -  0.4%  0-31
 3   Data_Pipe_to_memory     -  0.4%  0-31
 2    Floating_Point_Unit   0%  0.4%  0-31
 1     Integer_Pipeline     0%    0%  0-3
 4     Integer_Pipeline     0%    0%  4-7
 5     Integer_Pipeline     0%    0%  8-11
 6     Integer_Pipeline     0%  0.2%  12-15
 7     Integer_Pipeline     0%    0%  16-19
 8     Integer_Pipeline   2.8%  2.7%  20-23
 9     Integer_Pipeline   0.1%  0.2%  24-27
10     Integer_Pipeline     0%    0%  28-31
PG  RELATIONSHIP            HW    SW  CPUS
 0  System                   -  0.4%  0-31
 3   Data_Pipe_to_memory     -  0.4%  0-31
 2    Floating_Point_Unit   0%  0.4%  0-31
 1     Integer_Pipeline     0%  0.2%  0-3
 4     Integer_Pipeline     0%    0%  4-7
 5     Integer_Pipeline     0%    0%  8-11
 6     Integer_Pipeline     0%    0%  12-15
 7     Integer_Pipeline     0%    0%  16-19
 8     Integer_Pipeline   3.1%  2.5%  20-23
 9     Integer_Pipeline     0%    0%  24-27
10     Integer_Pipeline     0%  0.2%  28-31
Example 2 Displaying Information about Integer Pipeline

The following command displays detailed information about the two most utilized integer pipelines over the last two seconds.

$ pgstat -v -t 2 -r 'Integer_Pipeline' 1 2
PG  RELATIONSHIP         HW UTIL  CAP     SW    USR    SYS   IDLE CPUS
 1  Integer_Pipeline   0.2% 2.2M 1.4B   0.2%   0.0%   0.2%  99.8% 0-3
 4  Integer_Pipeline  13.1% 181M 1.4B  14.9%   0.0%  14.9%  85.1% 4-7
PG  RELATIONSHIP         HW UTIL  CAP     SW    USR    SYS   IDLE CPUS
 1  Integer_Pipeline   0.2% 1.9M 1.4B   0.2%   0.0%   0.2%  99.8% 0-3
 4  Integer_Pipeline  13.1% 163M 1.4B  14.9%   0.0%  14.9%  85.1% 4-7
Example 3 Displaying Core Utilization over Specified Period

The following command displays information about core utilization over the last two minutes.

$ pgstat -A 60 2
PG  RELATIONSHIP              HW      SW  CPUS
 0  System                     -   56.9%  0-31
 3   Data_Pipe_to_memory       -   56.9%  0-31
 2    Floating_Point_Unit   0.0%   56.9%  0-31
 1     Integer_Pipeline    36.7%   58.7%  0-3
 4     Integer_Pipeline    41.9%   58.3%  4-7
 5     Integer_Pipeline    31.0%   58.0%  8-11
 6     Integer_Pipeline    30.7%   57.9%  12-15
 7     Integer_Pipeline    30.1%   55.8%  16-19
 8     Integer_Pipeline    40.2%   54.8%  20-23
 9     Integer_Pipeline    35.0%   56.0%  24-27
10     Integer_Pipeline    40.3%   55.8%  28-31
PG  RELATIONSHIP              HW      SW  CPUS
 0  System                     -   10.7%  0-31
 3   Data_Pipe_to_memory       -   10.7%  0-31
 2    Floating_Point_Unit   0.0%   10.7%  0-31
 1     Integer_Pipeline     9.0%   10.7%  0-3
 4     Integer_Pipeline     9.6%   10.8%  4-7
 5     Integer_Pipeline     8.6%    9.9%  8-11
 6     Integer_Pipeline    10.5%   11.9%  12-15
 7     Integer_Pipeline     9.1%   10.4%  16-19
 8     Integer_Pipeline     9.6%   10.9%  20-23
 9     Integer_Pipeline     8.9%   10.0%  24-27
10     Integer_Pipeline     9.5%   10.7%  28-31

       SUMMARY: UTILIZATION OVER 120 SECONDS

                             ------HARDWARE------ ------SOFTWARE------
PG  RELATIONSHIP             MIN    AVG    MAX    MIN    AVG    MAX CPUS
 0  System                     -      -      -  10.7%  10.7%  56.9% 0-31
 3   Data_Pipe_to_memory       -      -      -  10.7%  10.7%  56.9% 0-31
 2    Floating_Point_Unit   0.0%   0.0%   0.0%  10.7%  10.7%  56.9% 0-31
 1     Integer_Pipeline     9.0%   8.5%  36.7%  10.7%  10.7%  58.7% 0-3
 4     Integer_Pipeline     9.6%   9.1%  41.9%  10.8%  10.8%  58.3% 4-7
 5     Integer_Pipeline     8.6%   8.1%  31.0%   9.9%   9.9%  58.0% 8-11
 6     Integer_Pipeline    10.5%  10.0%  30.7%  11.9%  11.9%  57.9% 12-15
 7     Integer_Pipeline     9.1%   8.6%  30.1%  10.4%  10.4%  55.8% 16-19
 8     Integer_Pipeline     9.6%   9.1%  40.2%  10.9%  10.9%  54.8% 20-23
 9     Integer_Pipeline     8.9%   8.4%  35.0%  10.0%  10.0%  56.0% 24-27
10     Integer_Pipeline     9.5%   8.9%  40.3%  10.7%  10.7%  55.8% 28-31
Example 4 Binning Hardware and Software Utilization by Chip

In the following example, the machine is a SPARC-T5 with 128 cores and 1024 strands.

Here we print a timestamp, followed by utilization aggregated by socket (chip).

# pgstat -T d -B chip
       July 11, 2013 02:53:11 PM PDT
       ID      RELATIONSHIP                            HW      SW  CPUS
           16  Chip (Software)                          -    0.2%  0-127
           16  Chip (Data_Pipe_to_memory)            0.1%       -  0-127
           16  Chip (Floating_Point_Unit)            0.0%       -  0-127
           16  Chip (Integer_Pipeline)               0.1%       -  0-127

        65552  Chip (Software)                          -    0.0%  128-255
        65552  Chip (Data_Pipe_to_memory)            0.0%       -  128-255
        65552  Chip (Floating_Point_Unit)            0.0%       -  128-255
        65552  Chip (Integer_Pipeline)               0.0%       -  128-255

       131088  Chip (Software)                          -    0.0%  256-383
       131088  Chip (Data_Pipe_to_memory)            0.0%       -  256-383
       131088  Chip (Floating_Point_Unit)            0.0%       -  256-383
       131088  Chip (Integer_Pipeline)               0.0%       -  256-383

       196624  Chip (Software)                          -    0.0%  384-511
       196624  Chip (Data_Pipe_to_memory)            0.0%       -  384-511
       196624  Chip (Floating_Point_Unit)            0.0%       -  384-511
       196624  Chip (Integer_Pipeline)               0.0%       -  384-511

       262160  Chip (Software)                          -    0.0%  512-639
       262160  Chip (Data_Pipe_to_memory)            0.0%       -  512-639
       262160  Chip (Floating_Point_Unit)            0.0%       -  512-639
       262160  Chip (Integer_Pipeline)               0.0%       -  512-639

       327696  Chip (Software)                          -    0.8%  640-767
       327696  Chip (Data_Pipe_to_memory)            0.9%       -  640-767
       327696  Chip (Floating_Point_Unit)            0.0%       -  640-767
       327696  Chip (Integer_Pipeline)               2.1%       -  640-767

       393232  Chip (Software)                          -    0.0%  768-895
       393232  Chip (Data_Pipe_to_memory)            0.0%       -  768-895
       393232  Chip (Floating_Point_Unit)            0.0%       -  768-895
       393232  Chip (Integer_Pipeline)               0.0%       -  768-895

       458768  Chip (Software)                          -    0.0%  896-1023
       458768  Chip (Data_Pipe_to_memory)            0.0%       -  896-1023
       458768  Chip (Floating_Point_Unit)            0.0%       -  896-1023
       458768  Chip (Integer_Pipeline)               0.0%       -  896-1023

Exit Status

The following exit values are returned:

0

Successful completion.

1

Unable to get PG information from the system.

2

Specified interval, count, or all CPUs, PGs, and sharing relationships invalid.

3

Invalid syntax.

Attributes

See attributes(7) for descriptions of the following attributes:

ATTRIBUTE TYPE
ATTRIBUTE VALUE
Availability
system/core-os
Interface Stability
Uncommitted

The command line options and output are Uncommitted.

See Also

cputrack(1), cpc(3CPC), libcpc(3LIB), attributes(7), attributes(7), dtrace(8), pginfo(8)

Notes

The stats which require access to hardware performance counters may not be available in certain virtual environments.