Investigate and mitigate the affect on block time a single validator can have

[cmwaters]

There's evidence in the mocha-3 network that a single validator (with less than a third of the voting power) can alter the block intervals of the blocks they produce by changing their local TimeoutCommit.

It is expected that at least 1/3 and as much as 2/3 (depending on how may signatures are present in the commit) of voting power should be required to alter block intervals. The fact that we can get a 3 second block when all but one validator has a TimeoutCommit set to 11 seconds is indicative that validators are continuing to vote on newly arrived proposals even while waiting for TimeoutCommit period from the previous height to pass.

It should be made mandatory that a node respect the full TimeoutCommit period before engaging in round 0 of the following height.

[evan-forbes]

can we close this?