Investigate parallelising cron.germinate by arch

[Adam Conrad]

It might be nice, assuming germinate isn't entirely I/O bound, to parallelise it a bit in the publisher. A quick discussion on IRC came to the conclusion that parallelising by flavour/arch combination wouldn't work so well, but parallelising purely per arch would probably work fine, and could cut the run time down a bit.

[Colin Watson (cjwatson)]

We can indeed safely parallelise by architecture. Experimentation with a cloned-and-hacked version of generate-extra-overrides on snakefruit, which has very similar specs to pepo, suggests that this takes us from 162 seconds to 55 seconds (median of five runs in both cases). That's about a 3x improvement rather than the nearly 6x (for six architectures) that we'd hoped for, but still not to be sneezed at. I'll see about making the corresponding changes to generate-extra-overrides.