Several people have brought this up (including @fgrzadkowski and @yujuhong). We have various performance/scalability goals (e.g. #4521, #3954, #3952) but we haven't stated what VM size we will use for the master when making these measurements. (I'm assuming we can just decide a GCE VM size, rather than a node configuration for every cloud provider we support.)

We've been setting GOMAXPROCS in various places (#4605, #4675). I'm not sure we've been very systematic about it, but either way, or performance presumably depends on # cores of the master. I believe people have mostly been using 4-core masters for the performance measurements? (I can't remember if we've been measuring memory consumption vs. cluster size, but if not, we should to ensure the VM has enough memory for a 100 node cluster).

Unless someone objects, I'm going to assume we are going to use 4-core VM.

@filipg @wojtek-t @satnam6502 @roberthbailey @lavalamp @bgrant0607 @brendandburns