Hi,
So, another option is that we modify Koji to submit sub-package tasks across multiple builders. i.e. all of the ARM subpackages (and this would happen for x86 variants too) would get submitted to builders at the same time, rather than linearly. It's a lot of work, but it's doable.
Hmm, that wouldn't be subpackages I guess, but different variants of the same source rpm, i.e. rpmbuild --define "variant={tegra,omap,...}" kernel.spec
Then have the specfile pick the kernel config + package name based on %variant.
Could be useful to parallelize other packages too:
qemu for example could be splitted by target arch, so you'll have variant={x86,arm,ppc,sparc,mips,...}. Wouldn't work as good as for the kernel though as there are common files for all archs, are compiled once with todays subpackages but would compiled multiple times with variants.
Maybe gcc could profit too? I remember there is (was?) a huge gcc build time difference depending on whenever java language support was enabled or not. So "./configure --languages=%{variant}" could possibly speed up gcc build time by building all langs in parallel?
cheers, Gerd