blob: 6fa9fe4fe8c489660346c2b1cd555748678be6ac (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
|
// Copyright (c) 2015 The Bitcoin Core developers
// Distributed under the MIT software license, see the accompanying
// file COPYING or http://www.opensource.org/licenses/mit-license.php.
#include "bench.h"
#include "util.h"
#include "validation.h"
#include "checkqueue.h"
#include "prevector.h"
#include <vector>
#include <boost/thread/thread.hpp>
#include "random.h"
// This Benchmark tests the CheckQueue with the lightest
// weight Checks, so it should make any lock contention
// particularly visible
static const int MIN_CORES = 2;
static const size_t BATCHES = 101;
static const size_t BATCH_SIZE = 30;
static const int PREVECTOR_SIZE = 28;
static const int QUEUE_BATCH_SIZE = 128;
static void CCheckQueueSpeed(benchmark::State& state)
{
struct FakeJobNoWork {
bool operator()()
{
return true;
}
void swap(FakeJobNoWork& x){};
};
CCheckQueue<FakeJobNoWork> queue {QUEUE_BATCH_SIZE};
boost::thread_group tg;
for (auto x = 0; x < std::max(MIN_CORES, GetNumCores()); ++x) {
tg.create_thread([&]{queue.Thread();});
}
while (state.KeepRunning()) {
CCheckQueueControl<FakeJobNoWork> control(&queue);
// We call Add a number of times to simulate the behavior of adding
// a block of transactions at once.
std::vector<std::vector<FakeJobNoWork>> vBatches(BATCHES);
for (auto& vChecks : vBatches) {
vChecks.resize(BATCH_SIZE);
}
for (auto& vChecks : vBatches) {
// We can't make vChecks in the inner loop because we want to measure
// the cost of getting the memory to each thread and we might get the same
// memory
control.Add(vChecks);
}
// control waits for completion by RAII, but
// it is done explicitly here for clarity
control.Wait();
}
tg.interrupt_all();
tg.join_all();
}
// This Benchmark tests the CheckQueue with a slightly realistic workload,
// where checks all contain a prevector that is indirect 50% of the time
// and there is a little bit of work done between calls to Add.
static void CCheckQueueSpeedPrevectorJob(benchmark::State& state)
{
struct PrevectorJob {
prevector<PREVECTOR_SIZE, uint8_t> p;
PrevectorJob(){
}
PrevectorJob(FastRandomContext& insecure_rand){
p.resize(insecure_rand.rand32() % (PREVECTOR_SIZE*2));
}
bool operator()()
{
return true;
}
void swap(PrevectorJob& x){p.swap(x.p);};
};
CCheckQueue<PrevectorJob> queue {QUEUE_BATCH_SIZE};
boost::thread_group tg;
for (auto x = 0; x < std::max(MIN_CORES, GetNumCores()); ++x) {
tg.create_thread([&]{queue.Thread();});
}
while (state.KeepRunning()) {
// Make insecure_rand here so that each iteration is identical.
FastRandomContext insecure_rand(true);
CCheckQueueControl<PrevectorJob> control(&queue);
std::vector<std::vector<PrevectorJob>> vBatches(BATCHES);
for (auto& vChecks : vBatches) {
vChecks.reserve(BATCH_SIZE);
for (size_t x = 0; x < BATCH_SIZE; ++x)
vChecks.emplace_back(insecure_rand);
control.Add(vChecks);
}
// control waits for completion by RAII, but
// it is done explicitly here for clarity
control.Wait();
}
tg.interrupt_all();
tg.join_all();
}
BENCHMARK(CCheckQueueSpeed);
BENCHMARK(CCheckQueueSpeedPrevectorJob);
|
