ColinIanKing

stress-ng vs sysbench stress-ng vs sysbench A practical guide to two common Linux load-generation tools: what each one does, how they differ, and when to use them. --- 1. Overview | | stress-ng | sysbench | |---|---|---| | Primary goal | Stress testing and fault finding | Benchmarking and performance measurement | | One-line summary | Designed to break things | Designed to measure things | --- 2. stress-ng stress-ng loads a system's hardware and kernel interfaces in many different ways. It is a much larger successor to the original stress tool. Core features - Large set of stressors: 300+ built-in tests, each exercising a specific part of the system. - CPU stress: Many compute methods (integer, floating point, matrix, FFT, bit operations, hashing) via --cpu and --cpu-method. - Memory stress: Virtual memory, cache, memory copy, mmap, page faults, NUMA, via --vm, --cache, --memcpy, --mmap. - I/O and filesystem stress: Disk reads/writes, fsync, file locking, directory operations, via --hdd, --io, --dir. - Kernel interface stress: Syscalls, signals, pipes, sockets, semaphores, message queues, futexes, threads, context switches, scheduler. - Network stress: Socket and protocol-level tests. - Device and driver stress: ioctl, /dev, /proc and /sys interfaces. Control and configuration | Option | Purpose | |---|---| | --cpu 4 / --cpu 0 | Number of workers (0 = all online CPUs) | | --timeout 60s | Stop after a set time | | --cpu-ops N | Stop after N operations | | --cpu-load 50 | Target a CPU utilization percentage | | --class cpu | Run a group of stressors (cpu, memory, io, scheduler, ...) | | --all, --random, --sequential | Permutation modes | | -x | Exclude specific stressors | Measurement and reporting - Bogo ops: A rough throughput measure, useful for same-machine comparison but not a formal benchmark. - Metrics: --metrics / --metrics-brief for summaries; --yaml or --log-file to save results. - Thermal and perf data: --thermalstat, --vmstat, --perf. - Verification: --verify checks that results are correct, helping catch hardware faults or memory errors. Common uses - Burn-in and stability testing of new hardware - Checking cooling and thermal throttling - Finding kernel bugs or race conditions - Testing behavior under resource pressure (CPU, RAM, I/O) Example commands # Stress all CPUs for 60 seconds with a summary stress-ng --cpu 0 --timeout 60s --metrics-brief # 4 CPU workers, 2 VM workers using 1GB each, 1 disk worker stress-ng --cpu 4 --vm 2 --vm-bytes 1G --hdd 1 --timeout 120s --metrics-brief # Run every CPU-class stressor in turn stress-ng --class cpu --sequential 0 --timeout 30s > Caution: stress-ng can make a system unresponsive, overheat, or crash, and some stressors can fill disks or trigger the OOM killer. Use it on test machines or with sensible limits, not carelessly on production. --- 3. sysbench sysbench is a scriptable benchmark tool that produces repeatable performance numbers for CPU, memory, disk, threads, and databases. Core features - CPU: Prime number computation benchmark. - Memory: Read/write throughput tests. - File I/O: Sequential and random read/write tests. - Threads and mutex: Scheduler and locking performance. - Database (OLTP): Workloads for MySQL and PostgreSQL. - Lua scripting: Write custom workloads. Example commands # CPU benchmark sysbench cpu --threads=4 --time=30 run # Memory throughput sysbench memory --threads=4 run # File I/O sysbench fileio --file-total-size=2G prepare sysbench fileio --file-total-size=2G --file-test-mode=rndrw run sysbench fileio --file-total-size=2G cleanup --- 4. Key Differences | Aspect | stress-ng | sysbench | |---|---|---| | Purpose | Stress testing, fault finding | Benchmarking, performance comparison | | Scope | Very broad (300+ stressors) | Narrow: CPU, memory, file I/O, threads, database | | Output | Bogo ops (rough, same-machine only) | Events/sec, latency percentiles, MiB/sec | | Database testing | None | Strong (OLTP for MySQL/PostgreSQL) | | Correctness checking | Yes (--verify) | No, measures speed only | | Extensibility | Modify the C source | Lua scripts | --- 5. Which One Should You Use? | Goal | Recommended tool | |---|---| | Burn-in, thermal, or stability testing | stress-ng | | Kernel bug hunting | stress-ng | | Comparing two servers or VM sizes | sysbench | | Measuring the effect of tuning changes | sysbench | | Database server performance | sysbench | Using them together The two tools complement each other: 1. Run stress-ng to confirm the system stays stable under heavy load. 2. Run sysbench to record how fast the system actually is. --- 6. Summary - stress-ng answers: "Does this system survive heavy load?" - sysbench answers: "How fast is this system?" References: - sysbench.org - ColinIanKing/stress-ng