I’m reviewing experiments and I am unable to reproduce the authors' speedup between two algorithms on the virtual machine they provided.

This is not the first time it happened and I'm starting to think that running the tests in a VM distorts the time ratios somehow. It shouldn't if you assume that it only runs x times slower than a real machine, but maybe it's more complicated than that…

@otini it is absolutely more complicated than that

@migratory But in what way? And could it explain inconsistencies in algorithmic speedup measurements?

@otini if they interact with the OS at all (using swap, networking, even memory allocation), absolutely; if their perf depends on the relative costs of arithmetic and memory access, VMs might change performance due to interactions with the host's memory management/paging or compete with the host OS for CPU cache

Sign in to participate in the conversation
Functional Café

The social network of the future: No ads, no corporate surveillance, ethical design, and decentralization! Own your data with Mastodon!