On Tue, Jul 23, 2013 at 12:27:03PM +0100, Tier Nolan wrote:
> I was thinking about a change to the rules for distinguishing between forks
> and maybe a BIP..

Please provide equations and data justifying the 'magic constants' in
this proposal.

Currently we do not relay blocks to peers if they conflict with blocks
in the best known chain. What changes exactly are you proposing to that
behavior?

> *Summary*
> 
> - Low POW headers should be broadcast by the network
> 
> If a header has more than 1/64 of the POW of a block, it should be
> broadcast.  This provides information on which fork is getting most of the
> hashing power.
> 
> - Miners should use the header information to decide on longest chain
> 
> The fork selection rule for miners should be biased towards staying on the
> fork that has the most hashing power.
> 
> This means that they might keep hashing on a fork that is 1-2 blocks
> shorter.
> 
> If most miners follow the rule, then it is the best strategy for other
> miners to also follow this rule.
> 
> - Advantages
> 
> This lowers the probability of natural and malicious reversals.
> 
> *Distributing low POW headers*
> 
> First block header messages that have more than 1/64 of the standard POW
> requirements would be forwarded.
> 
> This means the client needs to maintain a short term view of the entire
> header tree.
> 
> if (header extends header tree) {
>   if (header meets full POW) {
>     add to header tree;
>     forward to peers;
>     check if any blocks in storage now extend the header tree
>   } else {
>     if (header meets POW / 64) {
>       forward to peers;
>     }
> } else {
>   if (header meets POW) {
>     add to orphan header storage
>   }
> }
> 
> The storage could be limited and headers could be discarded after a while.
> 
> This has the extra advantage that it informs clients of forks that are
> receiving hashing power.
> 
> This could be linked to a protocol version to prevent disconnects due to
> invalid header messages.
> 
> *Determining the longest chain*
> 
> Each link would get extra credit for headers received.
> 
> Assume there are 2 forks starting at block A as the fork point.
> 
> A(63) <- B(72) <- C(37) <- D(58)
> 
> and
> 
> A(63) <- B'(6) <- C'(9) <- D'(4) <- E(7) <- F(6)
> 
> The numbers in brackets are the number of low POW headers received that
> have those blocks as parent.
> 
> The new rule is that the POW for a block is equal to
> 
> POW * (1 + (headers / 16))
> 
> Only headers within <some threshold> of the end of the (shorter) chain
> count.  However, in most cases, that doesn't matter since the fork point
> will act as the start point.  As long as miners keep headers for 30-40
> blocks, they will likely have all headers back to any reasonable fork point.
> 
> This means that the top fork is considered longer, since it has much more
> headers, even though it has 2 less blocks.
> 
> If 75% of the miners follow this rule, then the top fork will eventually
> catch up and win, so it is in the interests of the other 25% to follow the
> rule too.
> 
> Even if there isn't complete agreement on headers received, the fork that
> is getting the most hashing will naturally gain most of the headers, so
> ties will be broken quickly.

-- 
'peter'[:-1]@petertodd.org
000000000000001e1c3393788031c4e427b67cfd1b5e90a3b0de4fff094b2894