Silent Payments: Light Client Protocol

The study compared the storage requirements of different types of filters and tweaks applied over these blocks, providing significant insights into their efficiency and applicability.

The basic BIP-158 filter required 5.78 GB in total or roughly 2.84 MB per day, whereas the Taproot-only filter showed a dramatically reduced size, requiring only 0.33 GB in total or 0.14 MB per day. This indicates that the Taproot-only filter is approximately 17.63 times smaller than the basic filter overall. Specifically, it was 7.65 times smaller during the blocks from 800,000 to 849,999 and 23.61 times smaller over the last 10,000 blocks analyzed. Both independently written implementations matched exactly with Core's basic filters, affirming their accuracy and reliability.

Additional information on encoding methods was provided, noting the use of siphash-2-4 and Golomb-Rice coding (P=19, M=784931). It was revealed that the number of eligible taproot outputs was substantial, totaling 359,001,723, which carried 124,299,689 distinct x-only keys. Despite maintaining duplicates, the encoded data size for these outputs was only 0.958 GB, showing an efficiency of 6.02 times compared to the unfiltered data.

Further enhancements were considered through proposed "pruning tweaks" for transactions where all taproot UTXOs are spent, particularly addressing dust limits and cut-through strategies. For example, when no filter was applied, the restore payload was 15.08 GB. Implementing a dust limit of 546 reduced this to 13.70 GB, and combining cut-through with a dust limit further reduced the size significantly. The most notable reduction was observed with cut-through combined with a dust limit of 1000, plummeting the restore payload to just 0.48 GB or 3.2% of the original size.

These findings are detailed further in the source documentation available at GitHub, which provides deeper insights into the methodologies and results of these measurements.