Some comments:

Having said that, I don't have a strong feeling about the use of 20-percentile as threshold to increase the block size. That means the block size is increased only when most miners agree, which sounds ok to me.

However, using 20-percentile as threshold to DECREASE the block size could be very dangerous. Consider that the block size has been stable at 8MB for a few years. Everyone are happy with that. An attacker would just need to acquire 21% of mining power to break the status quo and send us all the way to 1MB. The only way to stop such attempt is to 51% attack the attacker. That'd be really ugly.

For technical and ethical reasons, I believe the thresholds for increase and decrease must be symmetrical: increase the block size when the x-percentile is bigger than the current size, decrease the block size when the (100-x)-percentile is smaller than the current size. The overall effect is: the block size remains unchanged unless 80% of miners agree to.

https://www.reddit.com/r/bitcoin_devlist/comments/3ekhg2/bip_draft_hardfork_bit_jl2012_at_xbthk_jul_23_2015/

https://github.com/jl2012/bips/blob/master/hardforkbit.mediawiki

  1. Replace static 1M block size hard limit with a floating limit ("hardLimit").
  2. hardLimit floats within the range 1-32M, inclusive.
  3. Initial value of hardLimit is 1M, preserving current system.
  4. Changing hardLimit is accomplished by encoding a proposed value within a block's coinbase scriptSig.
    1. Votes refer to a byte value, encoded within the pattern "/BV\d+/" Example: /BV8000000/ votes for 8,000,000 byte hardLimit. If there is more than one match with with pattern, the first match is counted.
    2. Absent/invalid votes and votes below minimum cap (1M) are counted as 1M votes. Votes above the maximum cap (32M) are counted as 32M votes.
    3. A new hardLimit is calculated at each difficult adjustment period (2016 blocks), and applies to the next 2016 blocks.
    4. Calculate hardLimit by examining the coinbase scriptSig votes of the previous 12,000 blocks, and taking the 20th percentile and 80th percentile.
    5. New hardLimit is the median of the followings:
      1. min(current hardLimit * 1.2, 20-percentile)
      2. max(current hardLimit / 1.2, 80-percentile)
      3. current hardLimit
  5. version 4 block: the coinbase of a version 4 block must match this pattern: "/BV\d+/"
  6. 70% rule: If 8,400 of the last 12,000 blocks are version 4 or greater, reject invalid version 4 blocks. (testnet4: 501 of last 1000)
  7. 80% rule ("Point of no return"): If 9,600 of the last 12,000 blocks are version 4 or greater, reject all version <= 3 blocks. (testnet4: 750 of last 1000)
  8. Block version number is calculated after masking out high 16 bits (final bit count TBD by versionBits outcome).
Jeff Garzik via bitcoin-dev 於 2015-09-02 23:33 寫到:
> BIP 100 initial public draft:
> https://github.com/jgarzik/bip100/blob/master/bip-0100.mediawiki [1]
> 
> Emphasis on "initial"  This is a starting point for the usual open
> source feedback/iteration cycle, not an endpoint that Must Be This
> Way.
> 
> 
> 
> Links:
> ------
> [1] https://github.com/jgarzik/bip100/blob/master/bip-0100.mediawiki
> 
> _______________________________________________
> bitcoin-dev mailing list
> bitcoin-dev@lists.linuxfoundation.org
> https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev