Skip to content

New Block Fee Estimation Algo

David Vorick requested to merge medians into master

Bunch of insight while writing up our new transaction fee mechanism for review.

  1. There's no point in considering the highest-value fees in each block when guessing the required fee. The ~lowest value fee in each block was enough to get in that block, so that's pretty much all you should be considering. Of course, lowest is a bit dangerous, so we grab the fee located in the 75%-ile for each block, meaning there are about 500kb of transactions with lower fees, and about 1.5mb of transactions with higher fees in the block. This makes malice harder, while also not getting you an unneeded value.

  2. Sia makes very heavy use of the child-pays-for-parent scheme. This means that for any set of transactions, the naive median fee is rather likely to be '0', which is not very helpful because it's not actually 0. So this PR includes a scheme that lumps transaction sets together and appropriately considers cp4p.

The result is a new block-based fee estimation that is way better than the one we merged yesterday.

Merge request reports