Ampere System Profiler

The Ampere System Profiler (ASP) is a system-level analysis profiler that can be used to look for common platform-level problems while running workloads. It also includes a visualizer in the form of Plotly graphs and tables in an HTML report. ASP can be used for workload characterization at a system level, a first step towards a balanced platform.

System pre-Requisites

Assumes user is running as root or had root privileges

Required Commands

• sar
• sensors
• route
• numastat
• perf
• python3

YUM Packages for RHEL based systems:

sudo yum install lm_sensors perf sysstat numactl net-tools python3

APT Packages for Ubuntu or Debian:

sudo apt install lm-sensors sysstat numactl net-tools linux-tools-common linux-tools-$(uname -r) python3

Python Modules:

sudo python3 -m pip install plotly pandas argparse

Output

ASP will generate a html format report under the root directory of ASP.
The report includes:

• CPU Utilization
• Per-Core Utilization
• CPU Power
• CPU Frequencies
• Disk I/O
• Network Utilization
• NUMA Stats
• Top CPU Hotspots

Example Report

Getting Started

1. Clone the Repository

2. Run Setup

source setup.sh

Building the Wheel

This project uses Poetry for packaging and dependency management.

Step 1: Build the Wheel

poetry build

This will generate a .whl file and .tar.gz source distribution in the dist/ directory.

Installing the Package Locally

After building the wheel:

pip install dist/system_metrics_collector-<version>-py3-none-any.whl

Replace with the actual version number shown in the wheel filename.

Usage

asp [OPTIONS]

Examples

Output Help Message

asp --help

Sample Commands

#Supported Collectors:
#cpu,cpu_freq,cpu_power,io,irq_affinity,network,numastat,perf

# generate 20 samples at 2 seconds interval, all collectors
asp -n 20 -i 2 -N $NETWORK_INTERFACE

# generate 20 samples at 1 seconds interval, only collect cpu utilization,cpu_freq and cpu_power
asp -n 20 -i 1 -c cpu,cpu_freq,cpu_power


# generate 20 samples at 2 seconds interval, all collectors execept perf 
asp -n 20 -i 2 -f


# generate 20 samples at 1 seconds interval, only collect cpu utilization, cpu_freq, io, network, and perf with lower frequency rate of 99
asp -n 20 -i 1 -c cpu,cpu_freq,io,network,perf -F 99

CLI Options

Option	Description
-n INTEGER	Number of samples to take
-i INTEGER	Interval in seconds between samples
-o TEXT	Output directory for collected data
-c TEXT	Comma-separated list of collectors to use
-f	Disable perf collection
-p	Disable plotting of results
-F INTEGER	Sampling frequency for perf collector (default: 4000 Hz)
-N TEXT	Network interface to monitor (e.g., eth0)
-h, --help	Show help message and exit

Output

• Data is stored as .dat files in the specified output directory.
• If plotting is enabled, an report.html file is generated after collection.

Tested on

• Ubuntu 20.04
• Ubuntu 22.04
• Ubuntu 24.04
• Ubuntu 25.04
• Fedora 40
• Fedora 41
• Fedora 42
• Debian 12
• Debian 13
• Oracle Linux 9

FAQ

• For non standard kernels, ensure perf is installed to see hotspots.
• For runs on the cloud, readings from Frequency and Perf Hotspots might not show up.
• Cloud runs will not show power.