pginfo & pgstat

A couple of very welcome commands crept into Solaris 11. They are pginfo and pgstat (these are links to the Oracle documentation site).

The two commands deal with "processor groups", this seems a bit of a misnomer to me as they are really about CPU topology. They report information and utilisation stats demonstrating the resource sharing going on on the system, and how threads are using those resources. It's probably easiest to use a couple of examples from the man pages to show this. First off pginfo:

$ pginfo -p -T
0 (System) CPUs: 0-31
`-- 3 (Data_Pipe_to_memory [system,chip]) CPUs: 0-31
    `-- 2 (Floating_Point_Unit [system,chip]) CPUs: 0-31
        |-- 1 (Integer_Pipeline [core]) CPUs: 0-3
        |-- 4 (Integer_Pipeline [core]) CPUs: 4-7
        |-- 5 (Integer_Pipeline [core]) CPUs: 8-11
        |-- 6 (Integer_Pipeline [core]) CPUs: 12-15
        |-- 7 (Integer_Pipeline [core]) CPUs: 16-19
        |-- 8 (Integer_Pipeline [core]) CPUs: 20-23
        |-- 9 (Integer_Pipeline [core]) CPUs: 24-27
        `-- 10 (Integer_Pipeline [core]) CPUs: 28-31

This shows a processor with 32 virtual CPUs, sharing a single floating point pipeline, 8 cores with a single integer pipe each - looks like an UltraSPARC T1 to me.

The output from pginfo shows the utilisation of the processor:

$ 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

It reports both software utilisation - meaning what work the operating system has assigned to the cores, plus it can report pipeline utilisation using the hardware counters. Pipeline utilisation indicates whether the core is saturated or not - each pipeline can be fully utilised before the core is running the maximal number of threads.

I'm pleased to see these tools appear. It is useful to have tools that report the topology of the system, and it is great to see tools that report actual hardware utilisation. On earlier releases of Solaris you can always use corestat to get similar data.