rocksdb/util/dynamic_bloom_test.cc
Peter Dillinger b55b2f45d0 Faster new DynamicBloom implementation (for memtable) (#5762)
Summary:
Since DynamicBloom is now only used in-memory, we're free to
change it without schema compatibility issues. The new implementation
is drawn from (with manifest permission)
303542a767/bloom_simulation_tests/foo.cc (L613)

This has several speed advantages over the prior implementation:
* Uses fastrange instead of %
* Minimum logic to determine first (and all) probed memory addresses
* (Major) Two probes per 64-bit memory fetch/write.
* Very fast and effective (murmur-like) hash expansion/re-mixing. (At
least on recent CPUs, integer multiplication is very cheap.)

While a Bloom filter with 512-bit cache locality has about a 1.15x FP
rate penalty (e.g. 0.84% to 0.97%), further restricting to two probes
per 64 bits incurs an additional 1.12x FP rate penalty (e.g. 0.97% to
1.09%). Nevertheless, the unit tests show no "mediocre" FP rate samples,
unlike the old implementation with more erratic FP rates.

Especially for the memtable, we expect speed to outweigh somewhat higher
FP rates. For example, a negative table query would have to be 1000x
slower than a BF query to justify doubling BF query time to shave 10% off
FP rate (working assumption around 1% FP rate). While that seems likely
for SSTs, my data suggests a speed factor of roughly 50x for the memtable
(vs. BF; ~1.5% lower write throughput when enabling memtable Bloom
filter, after this change).  Thus, it's probably not worth even 5% more
time in the Bloom filter to shave off 1/10th of the Bloom FP rate, or 0.1%
in absolute terms, and it's probably at least 20% slower to recoup that
much FP rate from this new implementation. Because of this, we do not see
a need for a 'locality' option that affects the MemTable Bloom filter
and have decoupled the MemTable Bloom filter from Options::bloom_locality.

Note that just 3% more memory to the Bloom filter (10.3 bits per key vs.
just 10) is able to make up for the ~12% FP rate drop in the new
implementation:

[] # Nearly "ideal" FP-wise but reasonably fast cache-local implementation
[~/wormhashing/bloom_simulation_tests] ./foo_gcc_IMPL_CACHE_WORM64_FROM32_any.out 10000000 6 10 $RANDOM 100000000
./foo_gcc_IMPL_CACHE_WORM64_FROM32_any.out time: 3.29372 sampled_fp_rate: 0.00985956 ...

[] # Close match to this new implementation
[~/wormhashing/bloom_simulation_tests] ./foo_gcc_IMPL_CACHE_MUL64_BLOCK_FROM32_any.out 10000000 6 10.3 $RANDOM 100000000
./foo_gcc_IMPL_CACHE_MUL64_BLOCK_FROM32_any.out time: 2.10072 sampled_fp_rate: 0.00985655 ...

[] # Old locality=1 implementation
[~/wormhashing/bloom_simulation_tests] ./foo_gcc_IMPL_CACHE_ROCKSDB_DYNAMIC_any.out 10000000 6 10 $RANDOM 100000000
./foo_gcc_IMPL_CACHE_ROCKSDB_DYNAMIC_any.out time: 3.95472 sampled_fp_rate: 0.00988943 ...

Also note the dramatic speed improvement vs. alternatives.

--

Performance unit test: DynamicBloomTest.concurrent_with_perf is updated
to report more precise timing data. (Measure running time of each
thread, not just longest running thread, etc.) Results averaged over
various sizes enabled with --enable_perf and 20 runs each; old dynamic
bloom refers to locality=1, the faster of the old:

old dynamic bloom, avg add latency = 65.6468
new dynamic bloom, avg add latency = 44.3809
old dynamic bloom, avg query latency = 50.6485
new dynamic bloom, avg query latency = 43.2186
old avg parallel add latency = 41.678
new avg parallel add latency = 24.5238
old avg parallel hit latency = 14.6322
new avg parallel hit latency = 12.3939
old avg parallel miss latency = 16.7289
new avg parallel miss latency = 12.2134

Tested on a dedicated 64-bit production machine at Facebook. Significant
improvement all around.

Despite now using std::atomic<uint64_t>, quick before-and-after test on
a 32-bit machine (Intel Atom N270, released 2008) shows no regression in
performance, in some cases modest improvement.

--

Performance integration test (synthetic): with DEBUG_LEVEL=0, used
TEST_TMPDIR=/dev/shm ./db_bench --benchmarks=fillrandom,readmissing,readrandom,stats --num=2000000
and optionally with -memtable_whole_key_filtering -memtable_bloom_size_ratio=0.01
300 runs each configuration.

Write throughput change by enabling memtable bloom:
Old locality=0: -3.06%
Old locality=1: -2.37%
New:            -1.50%
conclusion -> seems to substantially close the gap

Readmissing throughput change by enabling memtable bloom:
Old locality=0: +34.47%
Old locality=1: +34.80%
New:            +33.25%
conclusion -> maybe a small new penalty from FP rate

Readrandom throughput change by enabling memtable bloom:
Old locality=0: +31.54%
Old locality=1: +31.13%
New:            +30.60%
conclusion -> maybe also from FP rate (after memtable flush)

--

Another conclusion we can draw from this new implementation is that the
existing 32-bit hash function is not inherently crippling the Bloom
filter speed or accuracy, below about 5 million keys. For speed, the
implementation is essentially the same whether starting with 32-bits or
64-bits of hash; it just determines whether the first multiplication
after fastrange is a pseudorandom expansion or needed re-mix. Note that
this multiplication can occur while memory is fetching.

For accuracy, in a standard configuration, you need about 5 million
keys before you have about a 1.1x FP penalty due to using a
32-bit hash vs. 64-bit:

[~/wormhashing/bloom_simulation_tests] ./foo_gcc_IMPL_CACHE_MUL64_BLOCK_FROM32_any.out $((5 * 1000 * 1000 * 10)) 6 10 $RANDOM 100000000
./foo_gcc_IMPL_CACHE_MUL64_BLOCK_FROM32_any.out time: 2.52069 sampled_fp_rate: 0.0118267 ...
[~/wormhashing/bloom_simulation_tests] ./foo_gcc_IMPL_CACHE_MUL64_BLOCK_any.out $((5 * 1000 * 1000 * 10)) 6 10 $RANDOM 100000000
./foo_gcc_IMPL_CACHE_MUL64_BLOCK_any.out time: 2.43871 sampled_fp_rate: 0.0109059
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5762

Differential Revision: D17214194

Pulled By: pdillinger

fbshipit-source-id: ad9da031772e985fd6b62a0e1db8e81892520595
2019-09-05 14:59:25 -07:00

299 lines
8.4 KiB
C++

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
#ifndef GFLAGS
#include <cstdio>
int main() {
fprintf(stderr, "Please install gflags to run this test... Skipping...\n");
return 0;
}
#else
#include <cinttypes>
#include <algorithm>
#include <atomic>
#include <functional>
#include <memory>
#include <thread>
#include <vector>
#include "dynamic_bloom.h"
#include "logging/logging.h"
#include "memory/arena.h"
#include "port/port.h"
#include "test_util/testharness.h"
#include "test_util/testutil.h"
#include "util/gflags_compat.h"
#include "util/stop_watch.h"
using GFLAGS_NAMESPACE::ParseCommandLineFlags;
DEFINE_int32(bits_per_key, 10, "");
DEFINE_int32(num_probes, 6, "");
DEFINE_bool(enable_perf, false, "");
namespace rocksdb {
static Slice Key(uint64_t i, char* buffer) {
memcpy(buffer, &i, sizeof(i));
return Slice(buffer, sizeof(i));
}
class DynamicBloomTest : public testing::Test {};
TEST_F(DynamicBloomTest, EmptyFilter) {
Arena arena;
DynamicBloom bloom1(&arena, 100, 2);
ASSERT_TRUE(!bloom1.MayContain("hello"));
ASSERT_TRUE(!bloom1.MayContain("world"));
DynamicBloom bloom2(&arena, CACHE_LINE_SIZE * 8 * 2 - 1, 2);
ASSERT_TRUE(!bloom2.MayContain("hello"));
ASSERT_TRUE(!bloom2.MayContain("world"));
}
TEST_F(DynamicBloomTest, Small) {
Arena arena;
DynamicBloom bloom1(&arena, 100, 2);
bloom1.Add("hello");
bloom1.Add("world");
ASSERT_TRUE(bloom1.MayContain("hello"));
ASSERT_TRUE(bloom1.MayContain("world"));
ASSERT_TRUE(!bloom1.MayContain("x"));
ASSERT_TRUE(!bloom1.MayContain("foo"));
DynamicBloom bloom2(&arena, CACHE_LINE_SIZE * 8 * 2 - 1, 2);
bloom2.Add("hello");
bloom2.Add("world");
ASSERT_TRUE(bloom2.MayContain("hello"));
ASSERT_TRUE(bloom2.MayContain("world"));
ASSERT_TRUE(!bloom2.MayContain("x"));
ASSERT_TRUE(!bloom2.MayContain("foo"));
}
TEST_F(DynamicBloomTest, SmallConcurrentAdd) {
Arena arena;
DynamicBloom bloom1(&arena, 100, 2);
bloom1.AddConcurrently("hello");
bloom1.AddConcurrently("world");
ASSERT_TRUE(bloom1.MayContain("hello"));
ASSERT_TRUE(bloom1.MayContain("world"));
ASSERT_TRUE(!bloom1.MayContain("x"));
ASSERT_TRUE(!bloom1.MayContain("foo"));
DynamicBloom bloom2(&arena, CACHE_LINE_SIZE * 8 * 2 - 1, 2);
bloom2.AddConcurrently("hello");
bloom2.AddConcurrently("world");
ASSERT_TRUE(bloom2.MayContain("hello"));
ASSERT_TRUE(bloom2.MayContain("world"));
ASSERT_TRUE(!bloom2.MayContain("x"));
ASSERT_TRUE(!bloom2.MayContain("foo"));
}
static uint32_t NextNum(uint32_t num) {
if (num < 10) {
num += 1;
} else if (num < 100) {
num += 10;
} else if (num < 1000) {
num += 100;
} else {
num += 1000;
}
return num;
}
TEST_F(DynamicBloomTest, VaryingLengths) {
char buffer[sizeof(uint64_t)];
// Count number of filters that significantly exceed the false positive rate
int mediocre_filters = 0;
int good_filters = 0;
uint32_t num_probes = static_cast<uint32_t>(FLAGS_num_probes);
fprintf(stderr, "bits_per_key: %d num_probes: %d\n", FLAGS_bits_per_key,
num_probes);
for (uint32_t num = 1; num <= 10000; num = NextNum(num)) {
uint32_t bloom_bits = 0;
Arena arena;
bloom_bits = num * FLAGS_bits_per_key;
DynamicBloom bloom(&arena, bloom_bits, num_probes);
for (uint64_t i = 0; i < num; i++) {
bloom.Add(Key(i, buffer));
ASSERT_TRUE(bloom.MayContain(Key(i, buffer)));
}
// All added keys must match
for (uint64_t i = 0; i < num; i++) {
ASSERT_TRUE(bloom.MayContain(Key(i, buffer))) << "Num " << num
<< "; key " << i;
}
// Check false positive rate
int result = 0;
for (uint64_t i = 0; i < 10000; i++) {
if (bloom.MayContain(Key(i + 1000000000, buffer))) {
result++;
}
}
double rate = result / 10000.0;
fprintf(stderr,
"False positives: %5.2f%% @ num = %6u, bloom_bits = %6u\n",
rate * 100.0, num, bloom_bits);
if (rate > 0.0125)
mediocre_filters++; // Allowed, but not too often
else
good_filters++;
}
fprintf(stderr, "Filters: %d good, %d mediocre\n", good_filters,
mediocre_filters);
ASSERT_LE(mediocre_filters, good_filters / 5);
}
TEST_F(DynamicBloomTest, perf) {
StopWatchNano timer(Env::Default());
uint32_t num_probes = static_cast<uint32_t>(FLAGS_num_probes);
if (!FLAGS_enable_perf) {
return;
}
for (uint32_t m = 1; m <= 8; ++m) {
Arena arena;
const uint32_t num_keys = m * 8 * 1024 * 1024;
fprintf(stderr, "testing %" PRIu32 "M keys\n", m * 8);
DynamicBloom std_bloom(&arena, num_keys * 10, num_probes);
timer.Start();
for (uint64_t i = 1; i <= num_keys; ++i) {
std_bloom.Add(Slice(reinterpret_cast<const char*>(&i), 8));
}
uint64_t elapsed = timer.ElapsedNanos();
fprintf(stderr, "dynamic bloom, avg add latency %3g\n",
static_cast<double>(elapsed) / num_keys);
uint32_t count = 0;
timer.Start();
for (uint64_t i = 1; i <= num_keys; ++i) {
if (std_bloom.MayContain(Slice(reinterpret_cast<const char*>(&i), 8))) {
++count;
}
}
ASSERT_EQ(count, num_keys);
elapsed = timer.ElapsedNanos();
assert(count > 0);
fprintf(stderr, "dynamic bloom, avg query latency %3g\n",
static_cast<double>(elapsed) / count);
}
}
TEST_F(DynamicBloomTest, concurrent_with_perf) {
uint32_t num_probes = static_cast<uint32_t>(FLAGS_num_probes);
uint32_t m_limit = FLAGS_enable_perf ? 8 : 1;
uint32_t num_threads = 4;
std::vector<port::Thread> threads;
for (uint32_t m = 1; m <= m_limit; ++m) {
Arena arena;
const uint32_t num_keys = m * 8 * 1024 * 1024;
fprintf(stderr, "testing %" PRIu32 "M keys\n", m * 8);
DynamicBloom std_bloom(&arena, num_keys * 10, num_probes);
std::atomic<uint64_t> elapsed(0);
std::function<void(size_t)> adder([&](size_t t) {
StopWatchNano timer(Env::Default());
timer.Start();
for (uint64_t i = 1 + t; i <= num_keys; i += num_threads) {
std_bloom.AddConcurrently(
Slice(reinterpret_cast<const char*>(&i), 8));
}
elapsed += timer.ElapsedNanos();
});
for (size_t t = 0; t < num_threads; ++t) {
threads.emplace_back(adder, t);
}
while (threads.size() > 0) {
threads.back().join();
threads.pop_back();
}
fprintf(stderr, "dynamic bloom, avg parallel add latency %3g"
" nanos/key\n",
static_cast<double>(elapsed) / num_threads / num_keys);
elapsed = 0;
std::function<void(size_t)> hitter([&](size_t t) {
StopWatchNano timer(Env::Default());
timer.Start();
for (uint64_t i = 1 + t; i <= num_keys; i += num_threads) {
bool f =
std_bloom.MayContain(Slice(reinterpret_cast<const char*>(&i), 8));
ASSERT_TRUE(f);
}
elapsed += timer.ElapsedNanos();
});
for (size_t t = 0; t < num_threads; ++t) {
threads.emplace_back(hitter, t);
}
while (threads.size() > 0) {
threads.back().join();
threads.pop_back();
}
fprintf(stderr, "dynamic bloom, avg parallel hit latency %3g"
" nanos/key\n",
static_cast<double>(elapsed) / num_threads / num_keys);
elapsed = 0;
std::atomic<uint32_t> false_positives(0);
std::function<void(size_t)> misser([&](size_t t) {
StopWatchNano timer(Env::Default());
timer.Start();
for (uint64_t i = num_keys + 1 + t; i <= 2 * num_keys;
i += num_threads) {
bool f =
std_bloom.MayContain(Slice(reinterpret_cast<const char*>(&i), 8));
if (f) {
++false_positives;
}
}
elapsed += timer.ElapsedNanos();
});
for (size_t t = 0; t < num_threads; ++t) {
threads.emplace_back(misser, t);
}
while (threads.size() > 0) {
threads.back().join();
threads.pop_back();
}
fprintf(stderr, "dynamic bloom, avg parallel miss latency %3g"
" nanos/key, %f%% false positive rate\n",
static_cast<double>(elapsed) / num_threads / num_keys,
false_positives.load() * 100.0 / num_keys);
}
}
} // namespace rocksdb
int main(int argc, char** argv) {
::testing::InitGoogleTest(&argc, argv);
ParseCommandLineFlags(&argc, &argv, true);
return RUN_ALL_TESTS();
}
#endif // GFLAGS