Bonnie++ vs Quickbench

Bonnie++ is a free, command-line benchmark suite for measuring filesystem and disk I/O performance on Linux. It exercises sequential input and output, rewrites, random seeks, and file operations including creation, stat, and deletion of many small files. The project describes these workloads as useful for investigating I/O bottlenecks in I/O-intensive applications, including database work, and for simulating file activity associated with Squid or INN. Bonnie++ can test more than 2 GB of storage on a 32-bit machine. Its -D option enables direct I/O for bulk tests, while -b calls fsync after each write and syncs the directory after file creation or deletion. Semaphore support can synchronize multiple instances. Results can be emitted as plain text or CSV, with companion tools to convert CSV into HTML or formatted plain text. The project is licensed under GPL version 2.0, and the repository includes source code and Debian, RPM, and Sun-related packaging files. Its documentation warns against running the benchmark on live production machines because it can affect services and consume disk space or inodes.

Quickbench is a free Bash script for measuring basic performance on Linux-based hosts, including virtual machines. It reports CPU configuration and runs busy-loop benchmarks at different parallel job counts, returning loop rates. Memory checks report Linux-allocatable memory and run a simple bandwidth test that may reflect L1 or L2 cache performance. Storage checks use parallel dd transfers with 4k and 1M blocks, reporting bandwidth and I/O rates. They run on the filesystem in the current working directory; to measure another device, mount its filesystem there and run Quickbench from that location. The script is designed for minimal Linux servers and needs no specific dependencies. It takes one parameter: the maximum duration in seconds for each benchmark loop. The project recommends at least 30 seconds for statistically meaningful results and cautions that values below 10 seconds may be unreliable. CPU loop measurements are not calibrated, so comparisons call for matching Quickbench, kernel, Bash, and dd versions. The project describes storage testing as non-thorough and says it cannot measure I/O completion latency. It is licensed under GPL-3.0.