I imagine many startups with batch or non-time-critical processes do this sort of shaping naturally. It strikes me as a 'crontab' mentality: "What time should

I imagine many startups with batch or non-time-critical processes do this sort of shaping naturally. It strikes me as a 'crontab' mentality: "What time should I run job X every night? Whenever our load is lowest, and when nothing else is running." At Wikispaces we do this for many GB of nightly log parsing and data aggregation. We're also looking at running classes of non-critical notification jobs during nightly lulls. As for workload mixing, a good example in our environment is memcached and MogileFS. Memcached servers need as much RAM as possible and a bit of CPU. MogileFS storage nodes need as much disk as possible and a bit of CPU. They go great together.