There are several trace collection issues blocking the progress of this bug:

1. The traces collected from CT is no longer sliced into one temperature per trace. (Fixed by reintroduce story dependence)
2. Added temperature causes added time cost for trace computation, causing benchmark timeout. Computation costs 70% time during a page set, which is unnecessary, cuz we will use chrome-trace-processor locally to process the traces again. Trace collection costs 23% time, which is necessary, assuming that it's the time spent on uploading the trace data to the server. 
2.1 Ideally, we want to disable trace computation in CT to save 70% of time.
2.1.1 Disable metric calculation. (Fix, but doesn't reduce much time)
2.1.2 Disable model calculation, which is a major cost during trace computation.
2.2 Increase the threshold of the benchmark timeout. (Last resort)
2.3 Reduce benchmark retries from 3 to 1. Retries of a timeout benchmark are costly.
3. Trace collection timeout (Fixed by increase the threshold)
4. 12h timeout for each bot (Fixed by better sharding with newly introduced --max-page-per-bot in CT)
5. 60s timeout for navigate (Fixed by increase the threshold)
6. Exception "Neither CHROMIUM_OUTPUT_DIR nor CHROMIUM_OUT_DIR has been set" issue which causes failures and timeouts.

Thanks for the detailed list Max!

Let's break this out into a doc, which can be commented on by anyone with the link (under your chromium.org account).

That should make tracking progress here a bit easier.

Then link the doc here!

I've just organized the issues we currently have in this doc:

https://docs.google.com/a/chromium.org/document/d/1bCwF8gX0ti2fGqz8Y3dHVmiB9tIbGoHlRxA3M46shSE/edit?usp=sharing

Tim, do we want to close this bug? Do we sill want to get to "0.2: Investigate effect of service workers"?

Marking fixed SGTM.