From 171ca7745e77c9f78f26556457fe64e5b2004a75 Mon Sep 17 00:00:00 2001 From: Gavin Andresen Date: Mon, 17 Mar 2014 08:19:54 -0400 Subject: estimatefee / estimatepriority RPC methods New RPC methods: return an estimate of the fee (or priority) a transaction needs to be likely to confirm in a given number of blocks. Mike Hearn created the first version of this method for estimating fees. It works as follows: For transactions that took 1 to N (I picked N=25) blocks to confirm, keep N buckets with at most 100 entries in each recording the fees-per-kilobyte paid by those transactions. (separate buckets are kept for transactions that confirmed because they are high-priority) The buckets are filled as blocks are found, and are saved/restored in a new fee_estiamtes.dat file in the data directory. A few variations on Mike's initial scheme: To estimate the fee needed for a transaction to confirm in X buckets, all of the samples in all of the buckets are used and a median of all of the data is used to make the estimate. For example, imagine 25 buckets each containing the full 100 entries. Those 2,500 samples are sorted, and the estimate of the fee needed to confirm in the very next block is the 50'th-highest-fee-entry in that sorted list; the estimate of the fee needed to confirm in the next two blocks is the 150'th-highest-fee-entry, etc. That algorithm has the nice property that estimates of how much fee you need to pay to get confirmed in block N will always be greater than or equal to the estimate for block N+1. It would clearly be wrong to say "pay 11 uBTC and you'll get confirmed in 3 blocks, but pay 12 uBTC and it will take LONGER". A single block will not contribute more than 10 entries to any one bucket, so a single miner and a large block cannot overwhelm the estimates. --- src/txmempool.cpp | 370 ++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 370 insertions(+) (limited to 'src/txmempool.cpp') diff --git a/src/txmempool.cpp b/src/txmempool.cpp index 64c9eac73d..4bf01d4848 100644 --- a/src/txmempool.cpp +++ b/src/txmempool.cpp @@ -6,6 +6,8 @@ #include "core.h" #include "txmempool.h" +#include + using namespace std; CTxMemPoolEntry::CTxMemPoolEntry() @@ -35,12 +37,311 @@ CTxMemPoolEntry::GetPriority(unsigned int currentHeight) const return dResult; } +// +// Keep track of fee/priority for transactions confirmed within N blocks +// +class CBlockAverage +{ +private: + boost::circular_buffer feeSamples; + boost::circular_buffer prioritySamples; + + template std::vector buf2vec(boost::circular_buffer buf) const + { + std::vector vec(buf.begin(), buf.end()); + return vec; + } + +public: + CBlockAverage() : feeSamples(100), prioritySamples(100) { } + + void RecordFee(const CFeeRate& feeRate) { + feeSamples.push_back(feeRate); + } + + void RecordPriority(double priority) { + prioritySamples.push_back(priority); + } + + size_t FeeSamples() const { return feeSamples.size(); } + size_t GetFeeSamples(std::vector& insertInto) const + { + BOOST_FOREACH(const CFeeRate& f, feeSamples) + insertInto.push_back(f); + return feeSamples.size(); + } + size_t PrioritySamples() const { return prioritySamples.size(); } + size_t GetPrioritySamples(std::vector& insertInto) const + { + BOOST_FOREACH(double d, prioritySamples) + insertInto.push_back(d); + return prioritySamples.size(); + } + + // Used as belt-and-suspenders check when reading to detect + // file corruption + bool AreSane(const std::vector& vecFee) + { + BOOST_FOREACH(CFeeRate fee, vecFee) + { + if (fee < CFeeRate(0)) + return false; + if (fee.GetFee(1000) > CTransaction::minRelayTxFee.GetFee(1000) * 10000) + return false; + } + return true; + } + bool AreSane(const std::vector vecPriority) + { + BOOST_FOREACH(double priority, vecPriority) + { + if (priority < 0) + return false; + } + return true; + } + + void Write(CAutoFile& fileout) const + { + std::vector vecFee = buf2vec(feeSamples); + fileout << vecFee; + std::vector vecPriority = buf2vec(prioritySamples); + fileout << vecPriority; + } + + void Read(CAutoFile& filein) { + std::vector vecFee; + filein >> vecFee; + if (AreSane(vecFee)) + feeSamples.insert(feeSamples.end(), vecFee.begin(), vecFee.end()); + else + throw runtime_error("Corrupt fee value in estimates file."); + std::vector vecPriority; + filein >> vecPriority; + if (AreSane(vecPriority)) + prioritySamples.insert(prioritySamples.end(), vecPriority.begin(), vecPriority.end()); + else + throw runtime_error("Corrupt priority value in estimates file."); + if (feeSamples.size() + prioritySamples.size() > 0) + LogPrint("estimatefee", "Read %d fee samples and %d priority samples\n", + feeSamples.size(), prioritySamples.size()); + } +}; + +class CMinerPolicyEstimator +{ +private: + // Records observed averages transactions that confirmed within one block, two blocks, + // three blocks etc. + std::vector history; + std::vector sortedFeeSamples; + std::vector sortedPrioritySamples; + + int nBestSeenHeight; + + // nBlocksAgo is 0 based, i.e. transactions that confirmed in the highest seen block are + // nBlocksAgo == 0, transactions in the block before that are nBlocksAgo == 1 etc. + void seenTxConfirm(CFeeRate feeRate, double dPriority, int nBlocksAgo) + { + // Last entry records "everything else". + int nBlocksTruncated = min(nBlocksAgo, (int) history.size() - 1); + assert(nBlocksTruncated >= 0); + + // We need to guess why the transaction was included in a block-- either + // because it is high-priority or because it has sufficient fees. + bool sufficientFee = (feeRate > CTransaction::minRelayTxFee); + bool sufficientPriority = AllowFree(dPriority); + const char* assignedTo = "unassigned"; + if (sufficientFee && !sufficientPriority) + { + history[nBlocksTruncated].RecordFee(feeRate); + assignedTo = "fee"; + } + else if (sufficientPriority && !sufficientFee) + { + history[nBlocksTruncated].RecordPriority(dPriority); + assignedTo = "priority"; + } + else + { + // Neither or both fee and priority sufficient to get confirmed: + // don't know why they got confirmed. + } + LogPrint("estimatefee", "Seen TX confirm: %s : %s fee/%g priority, took %d blocks\n", + assignedTo, feeRate.ToString(), dPriority, nBlocksAgo); + } + +public: + CMinerPolicyEstimator(int nEntries) : nBestSeenHeight(0) + { + history.resize(nEntries); + } + + void seenBlock(const std::vector& entries, int nBlockHeight) + { + if (nBlockHeight <= nBestSeenHeight) + { + // Ignore side chains and re-orgs; assuming they are random + // they don't affect the estimate. + // And if an attacker can re-org the chain at will, then + // you've got much bigger problems than "attacker can influence + // transaction fees." + return; + } + nBestSeenHeight = nBlockHeight; + + // Fill up the history buckets based on how long transactions took + // to confirm. + std::vector > entriesByConfirmations; + entriesByConfirmations.resize(history.size()); + BOOST_FOREACH(const CTxMemPoolEntry& entry, entries) + { + // How many blocks did it take for miners to include this transaction? + int delta = nBlockHeight - entry.GetHeight(); + if (delta <= 0) + { + // Re-org made us lose height, this should only happen if we happen + // to re-org on a difficulty transition point: very rare! + continue; + } + if ((delta-1) >= (int)history.size()) + delta = history.size(); // Last bucket is catch-all + entriesByConfirmations[delta-1].push_back(&entry); + } + for (size_t i = 0; i < entriesByConfirmations.size(); i++) + { + std::vector &e = entriesByConfirmations.at(i); + // Insert at most 10 random entries per bucket, otherwise a single block + // can dominate an estimate: + if (e.size() > 10) { + std::random_shuffle(e.begin(), e.end()); + e.resize(10); + } + BOOST_FOREACH(const CTxMemPoolEntry* entry, e) + { + // Fees are stored and reported as BTC-per-kb: + CFeeRate feeRate(entry->GetFee(), entry->GetTxSize()); + double dPriority = entry->GetPriority(entry->GetHeight()); // Want priority when it went IN + seenTxConfirm(feeRate, dPriority, i); + } + } + for (size_t i = 0; i < history.size(); i++) { + if (history[i].FeeSamples() + history[i].PrioritySamples() > 0) + LogPrint("estimatefee", "estimates: for confirming within %d blocks based on %d/%d samples, fee=%s, prio=%g\n", + i, + history[i].FeeSamples(), history[i].PrioritySamples(), + estimateFee(i+1).ToString(), estimatePriority(i+1)); + } + sortedFeeSamples.clear(); + sortedPrioritySamples.clear(); + } + + // Can return CFeeRate(0) if we don't have any data for that many blocks back. nBlocksToConfirm is 1 based. + CFeeRate estimateFee(int nBlocksToConfirm) + { + nBlocksToConfirm--; + + if (nBlocksToConfirm < 0 || nBlocksToConfirm >= (int)history.size()) + return CFeeRate(0); + + if (sortedFeeSamples.size() == 0) + { + for (size_t i = 0; i < history.size(); i++) + history.at(i).GetFeeSamples(sortedFeeSamples); + std::sort(sortedFeeSamples.begin(), sortedFeeSamples.end(), + std::greater()); + } + if (sortedFeeSamples.size() == 0) + return CFeeRate(0); + + int nBucketSize = history.at(nBlocksToConfirm).FeeSamples(); + + // Estimates should not increase as number of confirmations goes up, + // but the estimates are noisy because confirmations happen discretely + // in blocks. To smooth out the estimates, use all samples in the history + // and use the nth highest where n is (number of samples in previous bucket + + // half the samples in nBlocksToConfirm bucket): + size_t nPrevSize = 0; + for (int i = 0; i < nBlocksToConfirm; i++) + nPrevSize += history.at(i).FeeSamples(); + size_t index = min(nPrevSize + nBucketSize/2, sortedFeeSamples.size()-1); + return sortedFeeSamples[index]; + } + double estimatePriority(int nBlocksToConfirm) + { + nBlocksToConfirm--; + + if (nBlocksToConfirm < 0 || nBlocksToConfirm >= (int)history.size()) + return -1; + + if (sortedPrioritySamples.size() == 0) + { + for (size_t i = 0; i < history.size(); i++) + history.at(i).GetPrioritySamples(sortedPrioritySamples); + std::sort(sortedPrioritySamples.begin(), sortedPrioritySamples.end(), + std::greater()); + } + if (sortedPrioritySamples.size() == 0) + return -1.0; + + int nBucketSize = history.at(nBlocksToConfirm).PrioritySamples(); + + // Estimates should not increase as number of confirmations needed goes up, + // but the estimates are noisy because confirmations happen discretely + // in blocks. To smooth out the estimates, use all samples in the history + // and use the nth highest where n is (number of samples in previous buckets + + // half the samples in nBlocksToConfirm bucket). + size_t nPrevSize = 0; + for (int i = 0; i < nBlocksToConfirm; i++) + nPrevSize += history.at(i).PrioritySamples(); + size_t index = min(nPrevSize + nBucketSize/2, sortedFeeSamples.size()-1); + return sortedPrioritySamples[index]; + } + + void Write(CAutoFile& fileout) const + { + fileout << nBestSeenHeight; + fileout << history.size(); + BOOST_FOREACH(const CBlockAverage& entry, history) + { + entry.Write(fileout); + } + } + + void Read(CAutoFile& filein) + { + filein >> nBestSeenHeight; + size_t numEntries; + filein >> numEntries; + history.clear(); + for (size_t i = 0; i < numEntries; i++) + { + CBlockAverage entry; + entry.Read(filein); + history.push_back(entry); + } + } +}; + + CTxMemPool::CTxMemPool() { // Sanity checks off by default for performance, because otherwise // accepting transactions becomes O(N^2) where N is the number // of transactions in the pool fSanityCheck = false; + + // 25 blocks is a compromise between using a lot of disk/memory and + // trying to give accurate estimates to people who might be willing + // to wait a day or two to save a fraction of a penny in fees. + // Confirmation times for very-low-fee transactions that take more + // than an hour or three to confirm are highly variable. + minerPolicyEstimator = new CMinerPolicyEstimator(25); +} + +CTxMemPool::~CTxMemPool() +{ + delete minerPolicyEstimator; } void CTxMemPool::pruneSpent(const uint256 &hashTx, CCoins &coins) @@ -128,6 +429,28 @@ void CTxMemPool::removeConflicts(const CTransaction &tx, std::list } } +// Called when a block is connected. Removes from mempool and updates the miner fee estimator. +void CTxMemPool::removeForBlock(const std::vector& vtx, unsigned int nBlockHeight, + std::list& conflicts) +{ + LOCK(cs); + std::vector entries; + BOOST_FOREACH(const CTransaction& tx, vtx) + { + uint256 hash = tx.GetHash(); + if (mapTx.count(hash)) + entries.push_back(mapTx[hash]); + } + minerPolicyEstimator->seenBlock(entries, nBlockHeight); + BOOST_FOREACH(const CTransaction& tx, vtx) + { + std::list dummy; + remove(tx, dummy, false); + removeConflicts(tx, conflicts); + } +} + + void CTxMemPool::clear() { LOCK(cs); @@ -195,6 +518,53 @@ bool CTxMemPool::lookup(uint256 hash, CTransaction& result) const return true; } +CFeeRate CTxMemPool::estimateFee(int nBlocks) const +{ + LOCK(cs); + return minerPolicyEstimator->estimateFee(nBlocks); +} +double CTxMemPool::estimatePriority(int nBlocks) const +{ + LOCK(cs); + return minerPolicyEstimator->estimatePriority(nBlocks); +} + +bool +CTxMemPool::WriteFeeEstimates(CAutoFile& fileout) const +{ + try { + LOCK(cs); + fileout << 99900; // version required to read: 0.9.99 or later + fileout << CLIENT_VERSION; // version that wrote the file + minerPolicyEstimator->Write(fileout); + } + catch (std::exception &e) { + LogPrintf("CTxMemPool::WriteFeeEstimates() : unable to write policy estimator data (non-fatal)"); + return false; + } + return true; +} + +bool +CTxMemPool::ReadFeeEstimates(CAutoFile& filein) +{ + try { + int nVersionRequired, nVersionThatWrote; + filein >> nVersionRequired >> nVersionThatWrote; + if (nVersionRequired > CLIENT_VERSION) + return error("CTxMemPool::ReadFeeEstimates() : up-version (%d) fee estimate file", nVersionRequired); + + LOCK(cs); + minerPolicyEstimator->Read(filein); + } + catch (std::exception &e) { + LogPrintf("CTxMemPool::ReadFeeEstimates() : unable to read policy estimator data (non-fatal)"); + return false; + } + return true; +} + + CCoinsViewMemPool::CCoinsViewMemPool(CCoinsView &baseIn, CTxMemPool &mempoolIn) : CCoinsViewBacked(baseIn), mempool(mempoolIn) { } bool CCoinsViewMemPool::GetCoins(const uint256 &txid, CCoins &coins) { -- cgit v1.2.3