Use half of cores available on machine.

[retrohacker]

Ceiling function for integer division

[rvagg]

fine by me, after you merge, trigger a docker hub rebuild, and wait for it to complete do this:

1. put the config entries into .ssh/config listed here: https://github.com/iojs/build/tree/master/setup/containers#readme
2. run this:

ssh ssh iojs-build-containers-1 docker pull -a iojs/build
ssh ssh iojs-build-containers-2 docker pull -a iojs/build

at the moment I think I'm favouring manual updates rather than setting up a hook for docker hub because it'll mean changing docker hub each time we privision new hosts and updating hopefully won't be a frequent exercise.

[retrohacker]

double ssh intentional?

[rvagg]

2 servers

[retrohacker]

Ah, I mean double on the same line: ssh ssh iojs-build...

[rvagg]

hah, no, not sure what that's about, just one will do

[bnoordhuis]

The assumption is that make is GNU make, right? And the intention of this PR is to not consume all cores in case there is more than one build running? Wouldn't it be better in that case to use -l <loadavg> instead of -j <jobs>? Something like make -l $(getconf _NPROCESSORS_ONLN) allows for burst-y behavior when the machine is otherwise unoccupied.

[rvagg]

@bnoordhuis I'd defer to you on the wisdom of this, I generally consider loadavg akin to voodoo but if you think this is a better approach then perhaps we should give it a go

[bnoordhuis]

I've had good results with it. The rule of thumb I use is that the requested load average should not be more than the number of processors. It's okay if those processors are hyperthreaded because compilation is normally memory bandwidth-constrained enough that it doesn't matter.