mpstat and pidstat - per-CPU and per-process statistics

mpstat and pidstat (from sysstat) are the per-resource performance tools. mpstat breaks down by CPU; pidstat breaks down by process. They are the next step after top.

mpstat

mpstat 1

Output:

Linux 6.1.0 (host)  2026-08-09  _x86_64_  (8 CPU)

14:30:00  CPU    %usr   %nice    %sys  %iowait    %irq   %soft  %steal  %guest  %idle
14:30:01  all   20.00   0.00    5.00      5.00    0.00   0.00     0.00    0.00  70.00
14:30:01    0   45.00   0.00    5.00      5.00    0.00   0.00     0.00    0.00  45.00
14:30:01    1   15.00   0.00    5.00      5.00    0.00   0.00     0.00    0.00  75.00
14:30:01    2   15.00   0.00    5.00      0.00    0.00   0.00     0.00    0.00  80.00
14:30:01    3   15.00   0.00    5.00      5.00    0.00   0.00     0.00    0.00  75.00
14:30:01    4   15.00   0.00    5.00      5.00    0.00   0.00     0.00    0.00  80.00
14:30:01    5   15.00   0.00    5.00      0.00    0.00   0.00     0.00    0.00  85.00
14:30:01    6   15.00   0.00    5.00      0.00    0.00   0.00     0.00    0.00  85.00
14:30:01    7   15.00   0.00    5.00      0.00    0.00   0.00     0.00    0.00  85.00

Reading:

• all: aggregated across all CPUs.
• CPU 0: 45% user, much higher than other CPUs. This is a single-threaded application bottleneck.
• CPU 0-3: similar %usr.
• %iowait: low (0-5%). Not I/O bound.

Single-thread usage on CPU 0 indicates a multi-thread application has a bottleneck on one core. Often this is locking or poor thread scaling.

pidstat

pidstat 1

Output:

14:30:00  UID    PID    %usr  %system   %guest  %wait  %CPU   CPU  Command
14:30:01  1000    1234   45.00    2.00     0.00   0.10  47.00    0  java
14:30:01  1000    1235    2.00    0.50     0.00   0.05   2.50    1  java
14:30:01  1000    1236    2.00    0.50     0.00   0.05   2.50    2  java
14:30:01  1000    1237    2.00    0.50     0.00   0.05   2.50    3  java
...

Each PID’s CPU usage. The application is multi-threaded, but one thread is using 47% CPU while others are at 2.5%. This is a thread imbalance - likely a lock contention or a single-threaded section in the code.

Context switches

pidstat -w 1

Output:

14:30:00  UID    PID   cswch/s  nvcswch/s  Command
14:30:01  1000    1234     500.00      1.50  java
14:30:01  1000    1235     800.00      2.10  java

cswch/s: voluntary context switches (process waits for I/O). nvcswch/s: involuntary (kernel preempts). High nvcswch/s indicates CPU saturation or thread contention.

Memory per-process

pidstat -r 1

Output:

14:30:00  UID    PID  minflt/s  majflt/s  VSZ   RSS   %MEM  Command
14:30:01  1000    1234      5.00      0.00  8G   2G   3.00  java

minflt/s: minor faults (no I/O). majflt/s: major faults (disk I/O). High majflt/s indicates memory pressure.

I/O per-process

pidstat -d 1

Output shows per-process read/write rates. The process doing the most I/O is the one to investigate.

Common patterns

Knowledge check