This week’s newsletter summarizes several talks from the Bitcoin Edge Dev++ training sessions and Scaling Bitcoin conference held last week in Tel Aviv. Also included is our regular section on notable changes to popular Bitcoin infrastructure projects.
None this week.
- Conference talk summaries: last week’s Scaling Bitcoin conference and Edge Dev++ training sessions yielded several interesting talks which we’ve summarized in two special sections later in this newsletter.
Bitcoin Edge Dev++
The Scaling Bitcoin conference was preceded by the community-organized Bitcoin Edge Dev++ training sessions. Recordings of the sessions are expected to be made public in the near future, but transcripts typed by Bryan Bishop are available now. We suggest at least skimming all of the topics, but we found the following transcripts both novel and interesting:
Bitcoin Core rebroadcasting logic by Amiti Uttarwar describes her work to eliminate a privacy leak in Bitcoin Core’s wallet. If the first send of a wallet transaction doesn’t result in reasonably fast confirmation, the wallet will rebroadcast the transaction in order to ensure it is relayed to miners. However, there’s no other case where a full node will rebroadcast a transaction it sent previously, so spy nodes can assume any node rebroadcasting a transaction is operated by the user who created that transaction. Worse, this behavior can be actively exploited by sending a tiny payment with a low fee to an address whose owner you want to identify and waiting for their wallet to rebroadcast the transaction.
Uttarwar’s proposed solution is having the node treat all transactions the same, rebroadcasting any of them when a heuristic indicates they should’ve been mined recently but weren’t. This prevents spy nodes from being able to assume that the node which rebroadcast a transaction is operated by the creator of that transaction. The presentation concluded with an overview of some edge cases, an insight from Uttarwar’s experience developing for Bitcoin Core, and a short list of open questions for future research. See Bitcoin Core #16698 for the first of Uttarwar’s PRs implementing these mitigations.
Blockchain design patterns: Layers and scaling approaches by Andrew Poelstra and David Vorick briefly describes a long list of existing and proposed technologies for making effective use of a space-limited block chain. Starting with existing features, they begin by comparing Bitcoin’s UTXO model to Ethereum’s balance model, finding that the spend-once nature of UTXOs greatly simplifies both security analysis and cache-based performance improvements. This effective caching is the basis of technologies such as bandwidth-reducing BIP152 compact blocks, latency-reducing FIBRE, and many CPU- and memory-reducing improvements within node software. Yet, Poelstra and Vorick note that the overall best way to reduce bandwidth, latency, CPU, and memory is to minimize the use of global state in the first place by looking for opportunities to use offchain protocols based on unbroadcast transactions, replacements of those unbroadcast transactions to allow state transitions, and hash locks to create dependencies between different transactions.
Looking at proposed technology, the presenters explain how the bandwidth overhead for relaying transactions grows linearly in the current protocol as you increase the number of your peers; this can be made almost constant with the proposed erlay protocol, allowing you to have many more peers, which reduces the risk of network partitioning attacks. Moving on to describing various parts of the taproot proposal, they show how schnorr signatures make it possible to validate multiple signatures at once (batch validation) and combine several public keys and signatures into a single pubkey and signature (signature aggregation), reducing the costs of general block validation and the specific costs for multisig users. Schnorr also makes it possible to create adaptor signatures that can provide the benefits of both a signature and a hash lock at the same time and for only the cost of one. Finally, taproot can commit to a set of conditions without requiring any party to reveal those conditions unless they need them, which they might not if their protocol allows them to use schnorr multiparty signatures instead. Each of these techniques, individually or in combination, can help users keep more data offchain.
The final part of their talk examines more speculative technology, such as improvements to LN using eltoo, the minimization of storage requirements using utreexo, client side validation which keeps almost all data offchain, Directed Acyclic Graph (DAG) based block chains that allow more frequent block production, confidential transactions that hide payment amounts, and sharding in both the form of federated block chains (available now) and other models that are more speculative.
Lightning network topology by Carla Kirk-Cohen describes the public topology of channels between LN nodes at present, what factors influenced that shape, and what we might be able to do to reshape it. She starts by noting the existence of private channels between some nodes that aren’t reflected in public data, and then outlines the current public network: “5600 nodes, about 35,218 public channels, and 959 public BTC sitting in those channels.” Many channels, including high-value channels, connect at one end to just a few popular nodes, such as the LNbig node which has about “20-24% of the liquidity of the LN in their channels.”
Looking at how we arrived at this hub-and-spoke type of topology, Kirk-Cohen notes the attractiveness of well-known nodes in the early network when all connections were made manually, then the attractiveness of well-connected nodes to the early LND autopilot, and finally the ongoing attractiveness of popular liquidity providers for people who want to receive or route payments.
Possible improvements discussed involve leveraging various graph metrics to help autopilots choose channels that are well-connected but not necessarily hubs (such as nodes that are one hop away from a hub). Additionally, she has been working on a scoring system for when to close a channel and so make your funds available for opening a channel to a more effective peer. Finally, she notes that changing channel connections costs onchain fees, so it’s not practical to make sudden changes to the topology; any changes are likely to be applied slowly over time as channels are opened and closed organically, so we may have to wait a long time to see what effect changes have.
The sixth Scaling Bitcoin conference was held in Tel Aviv, Israel, September 11th and 12th. Many of the topics and discussions focused on subjects we’ve already covered at length in the newsletter (such as Erlay, COSHV, signet, miniscript, and bitcoin vaults). Of the remaining topics transcribed by Bryan Bishop, we found the following topic particularly interesting:
TxProbe: discovering Bitcoin’s network topology using orphan transactions presented by Sergi Delgado-Segura (video, transcript, paper) describes the advantages and disadvantages of being able to determine Bitcoin’s network topology—which nodes connect to which other nodes—something Bitcoin Core tries to prevent. Delgado-Segura then discloses a novel technique he and his co-authors developed for probing the network to determine its topology, a combination of using orphan transactions (child transactions received before at least one of their parents) and a method for temporarily delaying the relay of transactions between peers. The talk illustrates some of the challenges of keeping Bitcoin secure and private at the P2P network level.
Although not described in detail during the talk, but perhaps worth mentioning here, are the mitigations Bitcoin Core developers have deployed as a result of the paper. The first of these was Bitcoin Core PR#14626, released as part of 0.18.0, which made orphan transactions less effective for probing (i.e., it fixes the problem described in section 4.3 of the TxProbe paper). PR#14897, described in Newsletter #33 with follow-up in Newsletters #43 and #51, provided a second and much more significant mitigation by eliminating the ability of a third-party to delay transaction propagation across the entire network (the invblock technique described in section 4.2 of the paper as well as a previously published paper by some of the same co-authors). A third mitigation was PR#15759, described in last week’s newsletter, which adds two outbound connections to each node that only relay blocks—making those two connections fundamentally resistant against transaction probing. These second and third mitigations are expected to be included in the upcoming Bitcoin Core 0.19 release.
Although several of the specific probing methods described in the presentation are being addressed, anyone interested in learning more about how nodes form a network and relay information should consider perusing the presentation or its paper.
Notable code and documentation changes
Bitcoin Core #16680 adds a
-chainconfiguration parameter that allows the user to select which block chain to use. This currently supports “main” (mainnet), “test” (testnet), and “regtest”. The later two chain names can currently be selected using the
-regtestparameters, but adding a generic function simplifies the introduction of additional chains later (e.g. using the signet code described in Newsletter #56).
Bitcoin Core #16787 extends the
getnetworkinfoRPCs with a new field that decodes the services bitfield which indicates what services the peer or local node offer. This is in addition to a previous field that provided the bitfield itself. For example, the bitfield
["NETWORK", "BLOOM", "WITNESS", "NETWORK_LIMITED"].
Bitcoin Core #16725 now omits the inferred
addressesfield from decoded transactions for P2PK outputs. P2PK outputs—payments directly to a pubkey—have never had an address format but the RPC interface previously returned the P2PKH address for that pubkey, confusing some users and developers.
Bitcoin Core #16714 updates the GUI first-time wizard with an option to enable block data pruning. The default value in the wizard depends on the available disk space, with pruning enabled by default if the user doesn’t have enough space to store the estimated block chain size plus 10 GB.
Bitcoin Core #16285 extends the
bestblockfields so that it’s possible to determine what chainstate was scanned.
Bitcoin Core #15584 disables support for the BIP70 payment protocol by default. It can be re-enabled by compiling with the
--enabled-bip70configure parameter. See our previous description in Newsletter #19 for several reasons why developers favor disabling BIP70 support.
LND #3485 removes the ability to upgrade from more than one major release in the past, simplifying the database migration code and allowing testing to focus on fewer upgrade paths. You can still upgrade older versions of LND by upgrading in steps (e.g., 0.5.2-beta → 0.6-beta → 0.7-beta → 0.7.1-beta).
C-Lightning #2945 makes new
sendpay_failurenotifications available to plugins.
C-Lightning #3010 implements experimental support for BOLT3
option_static_remotekey, allowing C-Lightning to honor the request from a channel peer to always pay the same public key for it rather than deriving a new key for it. This makes it easy for the remote peer to spend that money even if it loses some channel state. This option is only enabled if C-Lightning is compiled with experimental features enabled.