Code review comment for lp:~vanvugt/mir/predictive-bypass-2

@Alan's needs discussion

I think that measurements are becoming a more and more pressing need as we optimize the system further. We shouldn't increase latency once we have landed latency-reducing features, and we don't have a way to measure this thats well known.

Also, we're getting to the tricky part where our latency optimizations are dynamic... they're good in some cases, and not so good in others. A tricky problem like optimizing latency this really needs a tool to measure with a variety of interesting loads. This lets us see what tradeoffs we're making as we change the code, and guard against true regressions, where we accidentally increase latency accross-the-board inadventently.