|
| 1 | +// Copyright (c) The Bitcoin Core developers |
| 2 | +// Distributed under the MIT software license, see the accompanying |
| 3 | +// file COPYING or http://www.opensource.org/licenses/mit-license.php. |
| 4 | + |
| 5 | +#include <bench/bench.h> |
| 6 | + |
| 7 | +#include <util/bitset.h> |
| 8 | +#include <cluster_linearize.h> |
| 9 | + |
| 10 | +using namespace cluster_linearize; |
| 11 | + |
| 12 | +namespace { |
| 13 | + |
| 14 | +/** Construct a linear graph. These are pessimal for AncestorCandidateFinder, as they maximize |
| 15 | + * the number of ancestor set feerate updates. The best ancestor set is always the topmost |
| 16 | + * remaining transaction, whose removal requires updating all remaining transactions' ancestor |
| 17 | + * set feerates. */ |
| 18 | +template<typename SetType> |
| 19 | +DepGraph<SetType> MakeLinearGraph(ClusterIndex ntx) |
| 20 | +{ |
| 21 | + DepGraph<SetType> depgraph; |
| 22 | + for (ClusterIndex i = 0; i < ntx; ++i) { |
| 23 | + depgraph.AddTransaction({-int32_t(i), 1}); |
| 24 | + if (i > 0) depgraph.AddDependency(i - 1, i); |
| 25 | + } |
| 26 | + return depgraph; |
| 27 | +} |
| 28 | + |
| 29 | +// Construct a difficult graph. These need at least sqrt(2^(n-1)) iterations in the best |
| 30 | +// known algorithms (purely empirically determined). |
| 31 | +template<typename SetType> |
| 32 | +DepGraph<SetType> MakeHardGraph(ClusterIndex ntx) |
| 33 | +{ |
| 34 | + DepGraph<SetType> depgraph; |
| 35 | + for (ClusterIndex i = 0; i < ntx; ++i) { |
| 36 | + if (ntx & 1) { |
| 37 | + // Odd cluster size. |
| 38 | + // |
| 39 | + // Mermaid diagram code for the resulting cluster for 11 transactions: |
| 40 | + // ```mermaid |
| 41 | + // graph BT |
| 42 | + // T0["T0: 1/2"];T1["T1: 14/2"];T2["T2: 6/1"];T3["T3: 5/1"];T4["T4: 7/1"]; |
| 43 | + // T5["T5: 5/1"];T6["T6: 7/1"];T7["T7: 5/1"];T8["T8: 7/1"];T9["T9: 5/1"]; |
| 44 | + // T10["T10: 7/1"]; |
| 45 | + // T1-->T0;T1-->T2;T3-->T2;T4-->T3;T4-->T5;T6-->T5;T4-->T7;T8-->T7;T4-->T9;T10-->T9; |
| 46 | + // ``` |
| 47 | + if (i == 0) { |
| 48 | + depgraph.AddTransaction({1, 2}); |
| 49 | + } else if (i == 1) { |
| 50 | + depgraph.AddTransaction({14, 2}); |
| 51 | + depgraph.AddDependency(0, 1); |
| 52 | + } else if (i == 2) { |
| 53 | + depgraph.AddTransaction({6, 1}); |
| 54 | + depgraph.AddDependency(2, 1); |
| 55 | + } else if (i == 3) { |
| 56 | + depgraph.AddTransaction({5, 1}); |
| 57 | + depgraph.AddDependency(2, 3); |
| 58 | + } else if ((i & 1) == 0) { |
| 59 | + depgraph.AddTransaction({7, 1}); |
| 60 | + depgraph.AddDependency(i - 1, i); |
| 61 | + } else { |
| 62 | + depgraph.AddTransaction({5, 1}); |
| 63 | + depgraph.AddDependency(i, 4); |
| 64 | + } |
| 65 | + } else { |
| 66 | + // Even cluster size. |
| 67 | + // |
| 68 | + // Mermaid diagram code for the resulting cluster for 10 transactions: |
| 69 | + // ```mermaid |
| 70 | + // graph BT |
| 71 | + // T0["T0: 1"];T1["T1: 3"];T2["T2: 1"];T3["T3: 4"];T4["T4: 0"];T5["T5: 4"];T6["T6: 0"]; |
| 72 | + // T7["T7: 4"];T8["T8: 0"];T9["T9: 4"]; |
| 73 | + // T1-->T0;T2-->T0;T3-->T2;T3-->T4;T5-->T4;T3-->T6;T7-->T6;T3-->T8;T9-->T8; |
| 74 | + // ``` |
| 75 | + if (i == 0) { |
| 76 | + depgraph.AddTransaction({1, 1}); |
| 77 | + } else if (i == 1) { |
| 78 | + depgraph.AddTransaction({3, 1}); |
| 79 | + depgraph.AddDependency(0, 1); |
| 80 | + } else if (i == 2) { |
| 81 | + depgraph.AddTransaction({1, 1}); |
| 82 | + depgraph.AddDependency(0, 2); |
| 83 | + } else if (i & 1) { |
| 84 | + depgraph.AddTransaction({4, 1}); |
| 85 | + depgraph.AddDependency(i - 1, i); |
| 86 | + } else { |
| 87 | + depgraph.AddTransaction({0, 1}); |
| 88 | + depgraph.AddDependency(i, 3); |
| 89 | + } |
| 90 | + } |
| 91 | + } |
| 92 | + return depgraph; |
| 93 | +} |
| 94 | + |
| 95 | +/** Benchmark that does search-based candidate finding with 10000 iterations. |
| 96 | + * |
| 97 | + * Its goal is measuring how much time every additional search iteration in linearization costs. |
| 98 | + */ |
| 99 | +template<typename SetType> |
| 100 | +void BenchLinearizePerIterWorstCase(ClusterIndex ntx, benchmark::Bench& bench) |
| 101 | +{ |
| 102 | + const auto depgraph = MakeHardGraph<SetType>(ntx); |
| 103 | + const auto iter_limit = std::min<uint64_t>(10000, uint64_t{1} << (ntx / 2 - 1)); |
| 104 | + bench.batch(iter_limit).unit("iters").run([&] { |
| 105 | + SearchCandidateFinder finder(depgraph); |
| 106 | + auto [candidate, iters_performed] = finder.FindCandidateSet(iter_limit, {}); |
| 107 | + assert(iters_performed == iter_limit); |
| 108 | + }); |
| 109 | +} |
| 110 | + |
| 111 | +/** Benchmark for linearization of a trivial linear graph using just ancestor sort. |
| 112 | + * |
| 113 | + * Its goal is measuring how much time linearization may take without any search iterations. |
| 114 | + * |
| 115 | + * If P is the resulting time of BenchLinearizePerIterWorstCase, and N is the resulting time of |
| 116 | + * BenchLinearizeNoItersWorstCase, then an invocation of Linearize with max_iterations=m should |
| 117 | + * take no more than roughly N+m*P time. This may however be an overestimate, as the worst cases |
| 118 | + * do not coincide (the ones that are worst for linearization without any search happen to be ones |
| 119 | + * that do not need many search iterations). |
| 120 | + */ |
| 121 | +template<typename SetType> |
| 122 | +void BenchLinearizeNoItersWorstCase(ClusterIndex ntx, benchmark::Bench& bench) |
| 123 | +{ |
| 124 | + const auto depgraph = MakeLinearGraph<SetType>(ntx); |
| 125 | + bench.run([&] { |
| 126 | + Linearize(depgraph, /*max_iterations=*/0); |
| 127 | + }); |
| 128 | +} |
| 129 | + |
| 130 | +} // namespace |
| 131 | + |
| 132 | +static void LinearizePerIter16TxWorstCase(benchmark::Bench& bench) { BenchLinearizePerIterWorstCase<BitSet<16>>(16, bench); } |
| 133 | +static void LinearizePerIter32TxWorstCase(benchmark::Bench& bench) { BenchLinearizePerIterWorstCase<BitSet<32>>(32, bench); } |
| 134 | +static void LinearizePerIter48TxWorstCase(benchmark::Bench& bench) { BenchLinearizePerIterWorstCase<BitSet<48>>(48, bench); } |
| 135 | +static void LinearizePerIter64TxWorstCase(benchmark::Bench& bench) { BenchLinearizePerIterWorstCase<BitSet<64>>(64, bench); } |
| 136 | +static void LinearizePerIter75TxWorstCase(benchmark::Bench& bench) { BenchLinearizePerIterWorstCase<BitSet<75>>(75, bench); } |
| 137 | +static void LinearizePerIter99TxWorstCase(benchmark::Bench& bench) { BenchLinearizePerIterWorstCase<BitSet<99>>(99, bench); } |
| 138 | + |
| 139 | +static void LinearizeNoIters16TxWorstCase(benchmark::Bench& bench) { BenchLinearizeNoItersWorstCase<BitSet<16>>(16, bench); } |
| 140 | +static void LinearizeNoIters32TxWorstCase(benchmark::Bench& bench) { BenchLinearizeNoItersWorstCase<BitSet<32>>(32, bench); } |
| 141 | +static void LinearizeNoIters48TxWorstCase(benchmark::Bench& bench) { BenchLinearizeNoItersWorstCase<BitSet<48>>(48, bench); } |
| 142 | +static void LinearizeNoIters64TxWorstCase(benchmark::Bench& bench) { BenchLinearizeNoItersWorstCase<BitSet<64>>(64, bench); } |
| 143 | +static void LinearizeNoIters75TxWorstCase(benchmark::Bench& bench) { BenchLinearizeNoItersWorstCase<BitSet<75>>(75, bench); } |
| 144 | +static void LinearizeNoIters99TxWorstCase(benchmark::Bench& bench) { BenchLinearizeNoItersWorstCase<BitSet<99>>(99, bench); } |
| 145 | + |
| 146 | +BENCHMARK(LinearizePerIter16TxWorstCase, benchmark::PriorityLevel::HIGH); |
| 147 | +BENCHMARK(LinearizePerIter32TxWorstCase, benchmark::PriorityLevel::HIGH); |
| 148 | +BENCHMARK(LinearizePerIter48TxWorstCase, benchmark::PriorityLevel::HIGH); |
| 149 | +BENCHMARK(LinearizePerIter64TxWorstCase, benchmark::PriorityLevel::HIGH); |
| 150 | +BENCHMARK(LinearizePerIter75TxWorstCase, benchmark::PriorityLevel::HIGH); |
| 151 | +BENCHMARK(LinearizePerIter99TxWorstCase, benchmark::PriorityLevel::HIGH); |
| 152 | + |
| 153 | +BENCHMARK(LinearizeNoIters16TxWorstCase, benchmark::PriorityLevel::HIGH); |
| 154 | +BENCHMARK(LinearizeNoIters32TxWorstCase, benchmark::PriorityLevel::HIGH); |
| 155 | +BENCHMARK(LinearizeNoIters48TxWorstCase, benchmark::PriorityLevel::HIGH); |
| 156 | +BENCHMARK(LinearizeNoIters64TxWorstCase, benchmark::PriorityLevel::HIGH); |
| 157 | +BENCHMARK(LinearizeNoIters75TxWorstCase, benchmark::PriorityLevel::HIGH); |
| 158 | +BENCHMARK(LinearizeNoIters99TxWorstCase, benchmark::PriorityLevel::HIGH); |
0 commit comments