rtos / README.md
uncledecart's picture
Update README.md
858bd1a verified
metadata
license: apache-2.0

Content of this dataset

This dataset contains runs of experiments to evaluate feasibility of real-time capabilities of Dell R660XS(SN-CB5GXC4) server. There are 3 types of tests: caterpillar, cyclictest, 48-hour benchmark (for detailed explanation see below).

It has following tests runs

config folder test description
test1 05-19-35 caterpillar before BIOS correction
test2 06-38-03 cyclictest after BIOS correction
test3 15-12-05 48-hour with msr error
test4 10-49-18 48-hour wihout msr error
test5 11-04-24 48-hour using rdtset

For more info about analysis done check out this repo

What Is the Caterpillar Benchmark for RT?

The Caterpillar RT Benchmark is a diagnostic tool designed to:

  • Evaluate real-time responsiveness of CPU cores under varying system conditions.
  • Measure timing jitter (variations in scheduling and latency).
  • Identify hardware or kernel-level contributors to timing instability (e.g., IRQs, temperature fluctuations).

The benchmark runs multiple real-time workloads while collecting detailed monitoring data from the system.


Input Data Description

The benchmark generates several monitoring files in the output directory, each containing time-synchronized measurements from different subsystems.

File Description
output.csv Primary timing dataset (sample and session jitter metrics).
cpu_monitor.csv CPU temperature, frequency, and utilization per core.
cpustat_monitor.csv Detailed CPU statistics (user/system/idle times, context switches).
irq_monitor.csv IRQ activity counts per CPU core over time.
softirq_monitor.csv Kernel-level soft interrupt counts per CPU core.
meminfo_monitor.csv Memory pressure and allocation statistics.
sysinfo.json System configuration and metadata (CPU topology, kernel version, etc.).

Structure of output.csv

The file output.csv contains per-sample jitter measurements recorded during the benchmark.
Each row represents a single sampling point with both sample-level and session-level timing information.

Column Description Units
timestamp Timestamp of the measurement ISO 8601 datetime
SampleMin Minimum latency within current sampling window microseconds (µs)
SampleMax Maximum latency within current sampling window microseconds (µs)
SmplJitter Difference between SampleMax and SampleMin (per-sample jitter) microseconds (µs)
SessionMin Minimum latency observed since session start microseconds (µs)
SessionMax Maximum latency observed since session start microseconds (µs)
SessionJitter SessionMax - SessionMin, cumulative jitter for the session microseconds (µs)
Sample Sequential sample index integer

What Is the cyclictest Benchmark for RT?

Cyclictest repeatedly and precisely measures the gap between a thread’s scheduled wake-up time and the moment it actually wakes from clock_nanosleep, providing detailed latency statistics. It identifies real-time latency sources across hardware, firmware, and the operating system.

for more details see here

Input Data Description

The benchmark generates several monitoring files in the output directory, each containing time-synchronized measurements from different subsystems.

File Description
output.csv csv file containing test results
cpu_monitor.csv CPU temperature, frequency, and utilization per core.
cpustat_monitor.csv Detailed CPU statistics (user/system/idle times, context switches).
irq_monitor.csv IRQ activity counts per CPU core over time.
softirq_monitor.csv Kernel-level soft interrupt counts per CPU core.
meminfo_monitor.csv Memory pressure and allocation statistics.
sysinfo.json System configuration and metadata (CPU topology, kernel version, etc.).

Structure of output.csv

Short Explanation
timestamp Timestamp of the measurement
T Thread: Thread index and thread ID
P Priority: RT thread priority
I Interval: Intended wake up period for the latency measuring threads
C Count: Number of times the latency was measured (iteration count)
Min Minimum: Minimum latency that was measured
Act Actual: Latency measured during the latest completed iteration
Avg Average: Average latency that was measured
Max Maximum: Maximum latency that was measured