v5.0

This is the fifth release of this benchmark program. As always, the benchmark results of version 5.0 MUST NOT be compared to the results generated with any previous version.

Changelog:

• it is now possible to keep the result file - contribution by Enrico Falco
• added floating point oriented benchmark (haversine calculations)
• upgraded to Go 1.22.4
• updated dependencies

Version 4.0

This is the fourth release of this benchmark program. As always, the benchmark results of version 4.0 MUST NOT be compared to the results generated with any previous version.

Changelog:

• added logging benchmark simulating lines deduplication and PCI-DSS concealment
• added a new OLTP operating mode to study the evolution of CPU usage per throughput level
• upgraded to Go 1.20.2
• updated dependencies
• development machine is now a MacOS M1 Pro

Version 3.1

This is an update to previous version. The code of the benchmark is unchanged, but the compiler is now go 1.18.
This fixes a discrepancy between platforms (Intel/AMD and ARM) regarding the usage of the register-based ABI.
The register-based ABI improves the results on ARM64 by ~17%.

See this thread and this issue.

Version 3.0

This is the third release of this benchmark program.
Please note the benchmark results of version 3.0 MUST NOT be compared to the results generated with any previous version.

UPDATE: due to the dependency on Go 1.17, this version must not be used to compare the performance of x86_64 to Aarch64 CPU (i.e. ARM), because support for the register-based ABI has been added in Go 1.17 for x86_64, but only in Go 1.18 for Aarch64. The register-based ABI results in a significant performance boost (>10%), which is therefore unfair to Aarch64. See issue #8

Changelog:

• Adjusted memory/image benchmarks to reduce dependency on memmove
• Log benchmark version at runtime
• Added crypto benchmark based on 3DES/CTR (obsolete algorithms, but guaranteed without any assembly code)
• Upgraded to Go 1.17.4
• Upgraded dependencies
• Updated memory benchmark to ensure some preemption points
• Added Dijkstra's pearls algorithm benchmark
• Added simple route exploring benchmark with ad-hoc graph implementation
• Added a basic makefile to automate deliveries

Version 2.0

This is the second release of this benchmark program.
Please note the benchmark results of version 2.0 MUST NOT be compared to the results generated with previous version.

Changelog:

• Fixed stop of worker threads, improving the accuracy
• The default number of runs is now 10
• Added the -freq option to measure the CPU frequency
• Added new benchmarks: 8queens, memory, image
• Adjusted btree benchmark to favor CPU consumption over memory allocations
• Adjusted awk benchmark to reduce the weight of the second part
• Added statistic collection mechanism
• Minor cosmetic changes

v1.0

Version 1.0 - initial release