go-spacemesh
go-spacemesh copied to clipboard
spacemeshos
Allow tortoise active set size to decay more quickly
Description
Right now, if many miners disappear for any reason, self-healing ensures that consensus will eventually be re-established and tortoise will begin verifying layers again, but this may take a very long time. If it happens early in epoch N, then tortoise will be stuck for the rest of epoch N, as well as epoch N+1 (since most of the miners active in epoch N probably already submitted ATXs for N+1), and won't be able to heal until epoch N+2, when the size of the active set is finally reduced. With two-week epochs, this means that, in an extreme case, tortoise could be stuck for as long as a month.
Rather than waiting this long with a sudden cliff at the end of epoch N+1, we could cause the size of the active set (total miner weight) to decay exponentially, allowing the vote threshold to be crossed more quickly so that consensus can heal. This is a tradeoff between safety and liveness. It should decay gradually at first, then more quickly if the situation continues.
CC @tal-m
Affected code
Verifying tortoise, self-healing
Right now, if many miners disappear for any reason, self-healing ensures that consensus will eventually be re-established and tortoise will begin verifying layers again, but this may take a very long time.
if majority of miners disappear forever - self-healing will not ensure this. as there will be always shortage of weight, and global threshold won't be crossed.
there will be always shortage of weight, and global threshold won't be crossed
I don't think this is true. As I pointed out above, total weight will go down in subsequent epochs, and thresholds will eventually be crossed again (but it will take a while).
this issue can be transferred to research repository or to the forum research topic as the design isn't anywhere near ready for implementation
i will close it, if it comes up later in research discussions we will consider it again