Also covering BIP37 and Bloom filters
Transaction bloom filtering is a method that allows lightweight clients to limit the amount of transaction data they receive from full nodes to only those transactions that affect their wallet (plus a configurable amount of additional transactions to generate plausible deniability about which transactions belong to the client).
Bloom filters provide the ability to create a compact filter that is guaranteed to match a specified string with a configurable rate of false positive matches for other strings. A lightweight client can create a bloom filter for all of its wallet addresses, send that filter to a node using the P2P protocol messages defined in BIP37, and then request a special form of blocks (merkle blocks) from the node.
A merkle block, also defined in BIP37, will contain only transactions matching the previously sent filter plus the block header and a partial merkle branch connecting each matching transaction to the merkle root in the block header.
Clients will also receive announcements of new unconfirmed transactions being relayed by the node if they match the filter.
When BIP37 was popular, most lightweight clients that used it ran on mobile devices with limited bandwidth and so chose low false positive rates to minimize their bandwidth use. This meant that they essentially gave their list of addresses to any node they contacted. It was expected that privacy-focused users could mitigate this privacy loss by setting a higher false positive rate, but research suggests that the rate needs to be quite high in order to provide plausible deniability.
As an additional problem, nodes serving BIP37 filters must perform filtering independently for each client and it’s possible for filters to be created in a way that requires nodes perform an extensive amount of CPU processing to filter each block. This resulted in a set of known DoS vectors against nodes.
Although in practice BIP37 allowed clients to use a fairly small amount of bandwidth, it was slower and used more bandwidth than other remote transaction scanning methods based on large databases of transactions. Many popular lightweight clients today query such databases instead of using transaction bloom filters.
Note: this topic only refers to BIP37 transaction bloom filters. There are uses of generic bloom filters in Bitcoin (such as in Bitcoin Core’s transaction relay tracking) that aren’t indexed to this topic.
Primary code and documentation
Optech newsletter and website mentions
- Bitcoin Core 0.19 released with bloom filters disabled
- Bitcoin Core PR#16248 adds bloom filter whitelist option
- BRD field report: using native segwit addresses with bloom filters
- Mailing list discussion about disabling bloom filters in Bitcoin Core
- Bitcoin Core PR#16152 disables bloom filter support by default