Investigate/profile live migration performance

A short while ago I did an ad-hoc live migration for the first time in a while, and the memory transfer phase "felt" (how scientific!) a lot slower than it did the last time I tried one of these (which was probably a month or two ago).

In the short run, we should do some ad hoc tests to see if there are any obvious bottlenecks. (I'd like in particular to compare a few recent builds to see if I'm just imagining a speed difference.)

In the long run, we need better tools to assess blackout times and how they change over time. That probably includes

• Collecting information about at least the following:
  • how long a VM's vCPUs are paused in general
  • how long individual expensive phases of migration take, especially memory transfer and target device startup
• Surfacing this information so that we can get at it in test automation
• Writing automated tests that track migration performance over time
  • ex.: boot a guest; have it dirty as much memory as possible given its size (to try to get a consistent number of dirty pages to transfer); migrate it; gather migration metrics

These various steps should get their own issues; for now I'm just filing this to keep track of this work before I get distracted and forget about it.