forked from kohler/masstree-beta
-
Notifications
You must be signed in to change notification settings - Fork 0
/
perfstat.cc
216 lines (200 loc) · 8.01 KB
/
perfstat.cc
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
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
/* Masstree
* Eddie Kohler, Yandong Mao, Robert Morris
* Copyright (c) 2012-2013 President and Fellows of Harvard College
* Copyright (c) 2012-2013 Massachusetts Institute of Technology
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, subject to the conditions
* listed in the Masstree LICENSE file. These conditions include: you must
* preserve this copyright notice, and you cannot mention the copyright
* holders in advertising related to the Software without their permission.
* The Software is provided WITHOUT ANY WARRANTY, EXPRESS OR IMPLIED. This
* notice is a summary of the Masstree LICENSE file; the license in that file
* is legally binding.
*/
#include "perfstat.hh"
#include "compiler.hh"
#include "kvstats.hh"
#if HAVE_NUMA_H
#include <numa.h>
#endif
enum { MaxCores = 48 }; // Maximum number of cores kvdb statistics support
enum { MaxNumaNode = 8 }; // Maximum number of Numa node kvdb statistics support
enum { CoresPerChip = MaxCores / MaxNumaNode };
namespace Perf {
#if MEMSTATS && HAVE_NUMA_H && HAVE_LIBNUMA
static struct {
long long free;
long long size;
} numa[MaxNumaNode];
#endif
void
stat::initmain(bool pinthreads) {
(void) pinthreads;
#if PMC_ENABLED
always_assert(pinthreads && "Using performance counter requires pinning threads to cores!");
#endif
#if MEMSTATS && HAVE_NUMA_H && HAVE_LIBNUMA
if (numa_available() != -1) {
always_assert(numa_max_node() <= MaxNumaNode);
for (int i = 0; i <= numa_max_node(); i++)
numa[i].size = numa_node_size64(i, &numa[i].free);
}
#endif
}
template <typename T>
kvstats
sum_all_cores(const stat **s, int n, const int offset) {
kvstats sum;
for (int i = 0; i < n; i++) {
if (!s[i])
continue;
T v = *reinterpret_cast<const T *>(reinterpret_cast<const char *>(s[i]) + offset);
sum.add(v);
}
return sum;
}
template <typename T>
kvstats
sum_one_chip(const stat **s, int n, const int offset, const int chipidx) {
kvstats sum;
for (int i = 0; i < n; i++) {
if (!s[i] || s[i]->cid / (MaxCores / MaxNumaNode) != chipidx)
continue;
T v = *reinterpret_cast<const T *>(reinterpret_cast<const char *>(s[i]) + offset);
sum.add(v);
}
return sum;
}
template <typename T>
kvstats
sum_all_per_chip(const stat **s, int n, const int offset) {
kvstats per_chip[MaxNumaNode];
for (int i = 0; i < n; i++) {
if (!s[i])
continue;
T v = *reinterpret_cast<const T *>(reinterpret_cast<const char *>(s[i]) + offset);
per_chip[i / CoresPerChip].add(v);
}
kvstats sum;
for (int i = 0; i < MaxNumaNode; i++)
if (per_chip[i].count)
sum.add(per_chip[i].avg());
return sum;
}
void
stat::print(const stat **s, int n) {
(void)n;
(void)s;
#define sum_all_cores_of(field) \
sum_all_cores<typeof(s[0]->field)>(s, n, offsetof(Perf::stat, field))
#define sum_one_chip_of(field, c) \
sum_one_chip<typeof(s[0]->field)>(s, n, offsetof(Perf::stat, field), c)
#define sum_all_per_chip_of(field) \
sum_all_per_chip<typeof(s[0]->field)>(s, n, offsetof(Perf::stat, field))
#define sum_all_cores_of_array(field, oa) \
sum_all_cores<typeof(s[0]->field[0])>(s, n, offsetof(Perf::stat, field) + \
sizeof(s[0]->field[0]) * oa)
#define sum_one_chip_of_array(field, oa, c) \
sum_one_chip<typeof(s[0]->field[0])>(s, n, offsetof(Perf::stat, field) + \
sizeof(s[0]->field[0]) * oa, c)
#define sum_all_per_chip_of_array(field, oa) \
sum_all_per_chip<typeof(s[0]->field[0])>(s, n, offsetof(Perf::stat, field) + \
sizeof(s[0]->field[0]) * oa)
#if GETSTATS && 0
for (int i = 0; i < n; i++)
if (s[i]->ngets < 1000) {
s[i] = NULL;
continue;
}
kvstats ngets = sum_all_cores_of(ngets);
kvstats ntsc = sum_all_cores_of(ntsc);
kvstats np = sum_all_cores_of(nprobe);
if (np.sum >= 1)
fprintf(stderr, "Total probe %.0f, probe/get %.2f\n", np.sum, np.sum / ngets.sum);
#if PMC_ENABLED
fprintf(stderr, "(Inaccurate because PMC is Enabled!)");
#endif
fprintf(stderr, "Cycles/get (between mark_get_begin and mark_get_end): %.0f\n",
ntsc.sum / ngets.sum);
#if PMC_ENABLED
for (int i = 0; i < n; i++) {
if (!s[i])
continue;
fprintf(stderr, "Core %d:\n", i);
for (int pi = 0; pi < 4; pi++) {
fprintf(stderr, "\tpmc[%d]: %016" PRIx64 "->%016" PRIx64 "\n",
pi, s[i]->pmc_firstget[pi], s[i]->pmc_start[pi]);
always_assert(s[i]->pmc_start[pi] >= s[i]->pmc_firstget[pi]);
always_assert(s[i]->t1_lastget >= s[i]->t0_firstget);
}
}
// Compute the start and end time of get phase
kvstats getstart = sum_all_cores_of(t0_firstget);
kvstats getend = sum_all_cores_of(t1_lastget);
getstart.print_report("time of first get");
getend.print_report("time of last get");
// Compute per-chip pmc during the whole get phase
double pcpmc_phase[MaxNumaNode][4];
for (int i = 0; i < MaxNumaNode; i++)
for (int pi = 0; pi < 4; pi++)
pcpmc_phase[i][pi] = sum_one_chip_of_array(pmc_start, pi, i).avg() -
sum_one_chip_of_array(pmc_firstget, pi, i).avg();
// Compute cputime and realtime during get phase
kvstats t_firstget = sum_all_cores_of(t0_firstget);
kvstats t_lastget = sum_all_cores_of(t1_lastget);
double realtime = t_lastget.avg() - t_firstget.avg();
for (int pi = 0; pi < 4; pi++) {
fprintf(stderr, "DRAM access to node (pmc %d)\n", pi);
double sum = 0;
for (int i = 0; i < MaxNumaNode; i++) {
fprintf(stderr, "\tFrom chip %2d: %8.1f GB/s\n", i,
pcpmc_phase[i][pi] * 64 / (realtime * (1 << 30)));
sum += pcpmc_phase[i][pi];
}
fprintf(stderr, "\tSum: %8.1f GB/s\n",
sum * 64 / (realtime * (1 << 30)));
}
// Print per-get pmc_lookup
fprintf(stderr, "Per get statistics (counted between mark_get_begin and mark_get_end):\n");
for (int pi = 0; (ngets.sum > 0) && pi < 4; pi ++) {
kvstats pmc_lookup = sum_all_cores_of_array(pmc_lookup, pi);
kvstats pcpmc_lookup = sum_all_per_chip_of_array(pmc_lookup, pi);
fprintf(stderr, "\tpmc%d/get: %6.1f, per_chip_pmc%d/get: %6.1f\n",
pi, (double) pmc_lookup.sum / ngets.sum, pi,
(double) pcpmc_lookup.sum / ngets.sum);
}
#endif
#endif
#if MEMSTATS && HAVE_NUMA_H && HAVE_LIBNUMA && 0
// collect tree memory
kvstats tree_mem = sum_all_cores_of(tree_mem);
kvstats tree_keys = sum_all_cores_of(tree_keys);
fprintf(stderr, "Memory statistics\n");
fprintf(stderr, "\tAllocated per key: %.0f bytes, %.0f\n", tree_mem.sum / tree_keys.sum, tree_keys.sum);
if (numa_available() != -1) {
unsigned long total_alloc = 0;
for (int i = 0; i <= numa_max_node(); i++) {
kvstats chip = sum_one_chip_of(tree_mem, i);
long long nowfree;
long long size = numa_node_size64(i, &nowfree);
total_alloc += numa[i].free - nowfree;
fprintf(stderr, "\tNode %d (MB): size %6lld, allocated = %6lld - "
"%6lld = %6lld, tree_mem %6.0f\n",
i, size >> 20, numa[i].free >> 20, nowfree >> 20,
(numa[i].free - nowfree) / (1 << 20),
chip.sum / (1 << 20));
}
fprintf(stderr, "Total allocated memory %ld MB\n", total_alloc >> 20);
}
#endif
#if GCSTATS
// collect memory used by epoch based garbage collector
kvstats gc_nfree = sum_all_cores_of(gc_nfree);
kvstats gc_nalloc = sum_all_cores_of(gc_nalloc);
fprintf(stderr, "reuse per gc slot: %.0f, freed: %.0f, allocated: %.0f\n",
gc_nfree.sum / gc_nalloc.sum, gc_nfree.sum, gc_nalloc.sum);
#endif
}
}