Epoch 13 slots are now available and got enough slots that analysis could be made (via Minascan.io Blockberry API, thanks!). The result summary below. In brief, it works amazingly well, empty slots after the backfilling mechanism is introduced were less than 1.5 % of all slots, compared to original 45 %!
For the results, a minor change was introduced to earlier mechanism details: the requirement to remove a BP from the waitlist if it produced a block during its real slot was removed. Also, we can simplify the waitlist algorithm to be simple first-in, first-out (FIFO) mechanims with a limit of a unique BP public key in the waitlist, so SMT is not needed. SMT was originally thought to be used if Merkle paths of designated BP need to be added to consensus state.
Summary of Epoch 13: Slots 1528-7139 (5611 slots in total)
==========================================================
Total non-canonical blocks: 6761
Total empty slots: 84
Original empty slots: 2541
Waitlist time-window: 47
Waitlist length after last block: 10
Empty slots % without backfilling: 45.29 %
Empty slots % with backfilling: 1.50 %
Empty slots:
[1532, 1533, 1537, 1692, 3642, 3946, 3947, 3953, 3954, 3955, 3974, 4144, 4145, 4149, 4150, 4155, 4312, 4327, 4328, 4335, 4338, 4339, 4346, 4347, 4348, 4349, 4350, 4351, 4380, 4416, 4740, 5456, 5459, 5478, 5479, 5525, 5671, 5672, 5673, 5730, 5732, 5733, 5734, 5736, 5738, 5739, 5743, 5779, 5798, 5829, 5838, 5841, 5842, 5981, 5982, 5984, 5985, 5986, 5987, 5988, 6037, 6038, 6039, 6325, 6445, 6514, 6517, 6538, 6541, 6546, 6551, 6552, 6553, 6555, 6559, 6560, 6561, 6637, 6638, 6639, 6838, 6839, 6844, 6845]
AFAICT, the consensus state should have an extension of
- designated slot
- designated BP
- and designated slot proof
for the mechanism to work as analyzed (and described earlier).
All the data is available, so if you want to verify any parts of the analysis, I am happy to provide it! Just ping me in Discord for the data. Happy to also answer to questions here as well, of course!