Executive summary
Block time disruptions have lowered the rate of token production from the chain, and we're implementing a mechanism to shorten block times when this is the case, so that we catch up to our targeted production rates of 5 million HNT per month if we experience long block times.
Details
Currently, the lifetime average block interval is around 85 seconds, due to some long pauses, halts, and various growing pains. While we do our best to maintain good block times, we are ultimately running a complex distributed system on commodity networks and small machines running on mains power. Some level of disruption is inevitable.
Understandably, this variable rate of token emission is concerning, and we have put some thought
into various alternatives. While we considered time-based proration, we were concerned that the
incentives there were not particularly good, as anyone who wanted to minimize work while maximizing
return would simply do what they could to slow times down by interfering with the consensus group.
The strategy we have eventually settled on is a deterministic speedup. Typically, the next target
block time is calculated by taking the time of the last block and adding a chain variable called
block_time, which is currently 60000ms. If that time is in the past, we start immediately.
The current proposal, implemented in this PR, takes the lifetime average of block times by finding the time at block 187000 (a recent round number after the latest period of disruption), comparing it with the time from the last block, and nudging the target time back by the difference. Originally, we took a moving average over the last several hundred blocks, but we were concerned that this would be too expensive on the modest Hotspot hardware. This strategy, cribbed from NTP, means that we'll gently accelerate (and pleasingly, decelerate, should a bug cause us to produce blocks too quickly) until the average block time is on target.