amadeusitgroup/cpubench1a

cpubench1a is a CPU benchmark program suitable to evaluate the CPU capacity of physical or virtual machines.

cpubench1a's Issues

Test new CPU frequency measurement feature on ARM64

The CPU frequency measurement feature was tested on Intel/AMD CPUs but not on ARM64.

Workers stop mechanism is wrong

Workers are supposed to stop when they receive an Exit message. But the worker loop is written as:

	// Main worker loop, fetching operations from the input channel
	for op := range w.input {
		switch op {
		case OpStep:
			w.Step()
			w.nb++
		case OpExit:
			w.Exit()
			break
		default:
			log.Printf("Wrong operation %d", op)
		}
	}

The break statement applies to the switch instead of the loop. The benchmark stops anyway, but this results in possible inaccuracies in the way the time related to the last transactions is accounted.

Upgrade dependencies

One of the dependencies uses yaml.v3 which is now flagged as too old (security issue).
GoAWK now uses a bytecode VM, so the performance patterns may change.

Calibrate benchmark driver activity a bit better

The benchmark is running 1 driver with a buffered channel distributing the load to n workers.

As far as possible, we want to saturate the buffer as much as possible (until the very end of the benchmark), so that there is no wait state for the threads running the workers.

Reconsider the buffer capacity, the number of workers / thread and the batching level of the driver. Find a way to check there is no wait state.

Display version of the benchmark

cpubench1a should offer an option to display its version. The version should also be display in the output log.

Display aggregated results for -bench option

It would be useful to get a summary at the end of a -bench execution showing some statistics on the results of the iteration.
For instance, min, max, median and avg of single-threaded and multi-threaded iterations.

Consider a Windows port

The benchmark is clearly oriented towards server workloads, but we happen to have some Windows server boxes as well. Let's consider a Windows port for x86_64.

Setup CI/CD and static analysis

We have to setup Travis and some static analysis tool.

cpubench1a v3 unfair to non AMD64 CPUs

v3 uses go 1.17 which comes with a new way of passing function arguments and results using registers instead of the stack. However, this optimization was only implemented for AMD64.

Consequence: the v3 results when comparing Intel/AMD and ARM CPUs are biased.

go 1.18 will extend the new ABI to other 64 bits architectures including Aarch64 (i.e. ARM) for all OS. This should fix the bias.