2016-02-10 00:12:00 +01:00
|
|
|
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
// This source code is licensed under the BSD-style license found in the
|
|
|
|
// LICENSE file in the root directory of this source tree. An additional grant
|
|
|
|
// of patent rights can be found in the PATENTS file in the same directory.
|
2017-04-28 02:50:56 +02:00
|
|
|
// This source code is also licensed under the GPLv2 license found in the
|
|
|
|
// COPYING file in the root directory of this source tree.
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
//
|
|
|
|
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
|
|
|
|
// Use of this source code is governed by a BSD-style license that can be
|
|
|
|
// found in the LICENSE file. See the AUTHORS file for names of contributors.
|
|
|
|
|
|
|
|
#include <string>
|
|
|
|
#include "db/file_indexer.h"
|
|
|
|
#include "db/dbformat.h"
|
|
|
|
#include "db/version_edit.h"
|
2015-04-29 19:52:31 +02:00
|
|
|
#include "port/stack_trace.h"
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
#include "rocksdb/comparator.h"
|
|
|
|
#include "util/testharness.h"
|
|
|
|
#include "util/testutil.h"
|
|
|
|
|
|
|
|
namespace rocksdb {
|
|
|
|
|
|
|
|
class IntComparator : public Comparator {
|
|
|
|
public:
|
2015-02-26 20:28:41 +01:00
|
|
|
int Compare(const Slice& a, const Slice& b) const override {
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
assert(a.size() == 8);
|
|
|
|
assert(b.size() == 8);
|
2014-11-11 22:47:22 +01:00
|
|
|
int64_t diff = *reinterpret_cast<const int64_t*>(a.data()) -
|
|
|
|
*reinterpret_cast<const int64_t*>(b.data());
|
|
|
|
if (diff < 0) {
|
|
|
|
return -1;
|
|
|
|
} else if (diff == 0) {
|
|
|
|
return 0;
|
|
|
|
} else {
|
|
|
|
return 1;
|
|
|
|
}
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
}
|
|
|
|
|
2015-02-26 20:28:41 +01:00
|
|
|
const char* Name() const override { return "IntComparator"; }
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
|
2015-02-26 20:28:41 +01:00
|
|
|
void FindShortestSeparator(std::string* start,
|
|
|
|
const Slice& limit) const override {}
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
|
2015-02-26 20:28:41 +01:00
|
|
|
void FindShortSuccessor(std::string* key) const override {}
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
};
|
|
|
|
|
2015-03-17 22:08:00 +01:00
|
|
|
class FileIndexerTest : public testing::Test {
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
public:
|
2014-07-17 02:39:18 +02:00
|
|
|
FileIndexerTest()
|
|
|
|
: kNumLevels(4), files(new std::vector<FileMetaData*>[kNumLevels]) {}
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
|
|
|
|
~FileIndexerTest() {
|
2014-07-16 20:21:30 +02:00
|
|
|
ClearFiles();
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
delete[] files;
|
|
|
|
}
|
|
|
|
|
|
|
|
void AddFile(int level, int64_t smallest, int64_t largest) {
|
|
|
|
auto* f = new FileMetaData();
|
|
|
|
f->smallest = IntKey(smallest);
|
|
|
|
f->largest = IntKey(largest);
|
|
|
|
files[level].push_back(f);
|
|
|
|
}
|
|
|
|
|
|
|
|
InternalKey IntKey(int64_t v) {
|
|
|
|
return InternalKey(Slice(reinterpret_cast<char*>(&v), 8), 0, kTypeValue);
|
|
|
|
}
|
|
|
|
|
2014-07-16 20:21:30 +02:00
|
|
|
void ClearFiles() {
|
2014-04-29 21:33:57 +02:00
|
|
|
for (uint32_t i = 0; i < kNumLevels; ++i) {
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
for (auto* f : files[i]) {
|
|
|
|
delete f;
|
|
|
|
}
|
|
|
|
files[i].clear();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void GetNextLevelIndex(const uint32_t level, const uint32_t file_index,
|
|
|
|
const int cmp_smallest, const int cmp_largest, int32_t* left_index,
|
|
|
|
int32_t* right_index) {
|
|
|
|
*left_index = 100;
|
|
|
|
*right_index = 100;
|
2014-07-16 20:21:30 +02:00
|
|
|
indexer->GetNextLevelIndex(level, file_index, cmp_smallest, cmp_largest,
|
2014-07-17 02:39:18 +02:00
|
|
|
left_index, right_index);
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
}
|
|
|
|
|
2014-07-16 20:21:30 +02:00
|
|
|
int32_t left = 100;
|
|
|
|
int32_t right = 100;
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
const uint32_t kNumLevels;
|
|
|
|
IntComparator ucmp;
|
2014-07-16 20:21:30 +02:00
|
|
|
FileIndexer* indexer;
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
|
|
|
|
std::vector<FileMetaData*>* files;
|
|
|
|
};
|
|
|
|
|
2014-07-16 20:21:30 +02:00
|
|
|
// Case 0: Empty
|
2015-03-17 22:08:00 +01:00
|
|
|
TEST_F(FileIndexerTest, Empty) {
|
2014-07-16 20:21:30 +02:00
|
|
|
Arena arena;
|
|
|
|
indexer = new FileIndexer(&ucmp);
|
|
|
|
indexer->UpdateIndex(&arena, 0, files);
|
|
|
|
delete indexer;
|
|
|
|
}
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
|
2014-07-16 20:21:30 +02:00
|
|
|
// Case 1: no overlap, files are on the left of next level files
|
2015-03-17 22:08:00 +01:00
|
|
|
TEST_F(FileIndexerTest, no_overlap_left) {
|
2014-07-16 20:21:30 +02:00
|
|
|
Arena arena;
|
|
|
|
indexer = new FileIndexer(&ucmp);
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
// level 1
|
|
|
|
AddFile(1, 100, 200);
|
|
|
|
AddFile(1, 300, 400);
|
|
|
|
AddFile(1, 500, 600);
|
|
|
|
// level 2
|
|
|
|
AddFile(2, 1500, 1600);
|
|
|
|
AddFile(2, 1601, 1699);
|
|
|
|
AddFile(2, 1700, 1800);
|
|
|
|
// level 3
|
|
|
|
AddFile(3, 2500, 2600);
|
|
|
|
AddFile(3, 2601, 2699);
|
|
|
|
AddFile(3, 2700, 2800);
|
2014-07-16 20:21:30 +02:00
|
|
|
indexer->UpdateIndex(&arena, kNumLevels, files);
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
for (uint32_t level = 1; level < 3; ++level) {
|
|
|
|
for (uint32_t f = 0; f < 3; ++f) {
|
|
|
|
GetNextLevelIndex(level, f, -1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(0, left);
|
|
|
|
ASSERT_EQ(-1, right);
|
|
|
|
GetNextLevelIndex(level, f, 0, -1, &left, &right);
|
|
|
|
ASSERT_EQ(0, left);
|
|
|
|
ASSERT_EQ(-1, right);
|
|
|
|
GetNextLevelIndex(level, f, 1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(0, left);
|
|
|
|
ASSERT_EQ(-1, right);
|
|
|
|
GetNextLevelIndex(level, f, 1, 0, &left, &right);
|
|
|
|
ASSERT_EQ(0, left);
|
|
|
|
ASSERT_EQ(-1, right);
|
|
|
|
GetNextLevelIndex(level, f, 1, 1, &left, &right);
|
|
|
|
ASSERT_EQ(0, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
}
|
|
|
|
}
|
2014-07-16 20:21:30 +02:00
|
|
|
delete indexer;
|
|
|
|
ClearFiles();
|
|
|
|
}
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
|
2014-07-16 20:21:30 +02:00
|
|
|
// Case 2: no overlap, files are on the right of next level files
|
2015-03-17 22:08:00 +01:00
|
|
|
TEST_F(FileIndexerTest, no_overlap_right) {
|
2014-07-16 20:21:30 +02:00
|
|
|
Arena arena;
|
|
|
|
indexer = new FileIndexer(&ucmp);
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
// level 1
|
|
|
|
AddFile(1, 2100, 2200);
|
|
|
|
AddFile(1, 2300, 2400);
|
|
|
|
AddFile(1, 2500, 2600);
|
|
|
|
// level 2
|
|
|
|
AddFile(2, 1500, 1600);
|
|
|
|
AddFile(2, 1501, 1699);
|
|
|
|
AddFile(2, 1700, 1800);
|
|
|
|
// level 3
|
|
|
|
AddFile(3, 500, 600);
|
|
|
|
AddFile(3, 501, 699);
|
|
|
|
AddFile(3, 700, 800);
|
2014-07-16 20:21:30 +02:00
|
|
|
indexer->UpdateIndex(&arena, kNumLevels, files);
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
for (uint32_t level = 1; level < 3; ++level) {
|
|
|
|
for (uint32_t f = 0; f < 3; ++f) {
|
|
|
|
GetNextLevelIndex(level, f, -1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(f == 0 ? 0 : 3, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
GetNextLevelIndex(level, f, 0, -1, &left, &right);
|
|
|
|
ASSERT_EQ(3, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
GetNextLevelIndex(level, f, 1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(3, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
GetNextLevelIndex(level, f, 1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(3, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
GetNextLevelIndex(level, f, 1, 0, &left, &right);
|
|
|
|
ASSERT_EQ(3, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
GetNextLevelIndex(level, f, 1, 1, &left, &right);
|
|
|
|
ASSERT_EQ(3, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
}
|
|
|
|
}
|
2014-07-16 20:21:30 +02:00
|
|
|
delete indexer;
|
|
|
|
}
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
|
2014-07-16 20:21:30 +02:00
|
|
|
// Case 3: empty L2
|
2015-03-17 22:08:00 +01:00
|
|
|
TEST_F(FileIndexerTest, empty_L2) {
|
2014-07-16 20:21:30 +02:00
|
|
|
Arena arena;
|
|
|
|
indexer = new FileIndexer(&ucmp);
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
for (uint32_t i = 1; i < kNumLevels; ++i) {
|
2014-07-16 20:21:30 +02:00
|
|
|
ASSERT_EQ(0U, indexer->LevelIndexSize(i));
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
}
|
|
|
|
// level 1
|
|
|
|
AddFile(1, 2100, 2200);
|
|
|
|
AddFile(1, 2300, 2400);
|
|
|
|
AddFile(1, 2500, 2600);
|
|
|
|
// level 3
|
|
|
|
AddFile(3, 500, 600);
|
|
|
|
AddFile(3, 501, 699);
|
|
|
|
AddFile(3, 700, 800);
|
2014-07-16 20:21:30 +02:00
|
|
|
indexer->UpdateIndex(&arena, kNumLevels, files);
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
for (uint32_t f = 0; f < 3; ++f) {
|
|
|
|
GetNextLevelIndex(1, f, -1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(0, left);
|
|
|
|
ASSERT_EQ(-1, right);
|
|
|
|
GetNextLevelIndex(1, f, 0, -1, &left, &right);
|
|
|
|
ASSERT_EQ(0, left);
|
|
|
|
ASSERT_EQ(-1, right);
|
|
|
|
GetNextLevelIndex(1, f, 1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(0, left);
|
|
|
|
ASSERT_EQ(-1, right);
|
|
|
|
GetNextLevelIndex(1, f, 1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(0, left);
|
|
|
|
ASSERT_EQ(-1, right);
|
|
|
|
GetNextLevelIndex(1, f, 1, 0, &left, &right);
|
|
|
|
ASSERT_EQ(0, left);
|
|
|
|
ASSERT_EQ(-1, right);
|
|
|
|
GetNextLevelIndex(1, f, 1, 1, &left, &right);
|
|
|
|
ASSERT_EQ(0, left);
|
|
|
|
ASSERT_EQ(-1, right);
|
|
|
|
}
|
2014-07-16 20:21:30 +02:00
|
|
|
delete indexer;
|
|
|
|
ClearFiles();
|
|
|
|
}
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
|
2014-07-16 20:21:30 +02:00
|
|
|
// Case 4: mixed
|
2015-03-17 22:08:00 +01:00
|
|
|
TEST_F(FileIndexerTest, mixed) {
|
2014-07-16 20:21:30 +02:00
|
|
|
Arena arena;
|
|
|
|
indexer = new FileIndexer(&ucmp);
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
// level 1
|
|
|
|
AddFile(1, 100, 200);
|
|
|
|
AddFile(1, 250, 400);
|
|
|
|
AddFile(1, 450, 500);
|
|
|
|
// level 2
|
|
|
|
AddFile(2, 100, 150); // 0
|
|
|
|
AddFile(2, 200, 250); // 1
|
|
|
|
AddFile(2, 251, 300); // 2
|
|
|
|
AddFile(2, 301, 350); // 3
|
|
|
|
AddFile(2, 500, 600); // 4
|
|
|
|
// level 3
|
|
|
|
AddFile(3, 0, 50);
|
|
|
|
AddFile(3, 100, 200);
|
|
|
|
AddFile(3, 201, 250);
|
2014-07-16 20:21:30 +02:00
|
|
|
indexer->UpdateIndex(&arena, kNumLevels, files);
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
// level 1, 0
|
|
|
|
GetNextLevelIndex(1, 0, -1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(0, left);
|
|
|
|
ASSERT_EQ(0, right);
|
|
|
|
GetNextLevelIndex(1, 0, 0, -1, &left, &right);
|
|
|
|
ASSERT_EQ(0, left);
|
|
|
|
ASSERT_EQ(0, right);
|
|
|
|
GetNextLevelIndex(1, 0, 1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(0, left);
|
|
|
|
ASSERT_EQ(1, right);
|
|
|
|
GetNextLevelIndex(1, 0, 1, 0, &left, &right);
|
|
|
|
ASSERT_EQ(1, left);
|
|
|
|
ASSERT_EQ(1, right);
|
|
|
|
GetNextLevelIndex(1, 0, 1, 1, &left, &right);
|
|
|
|
ASSERT_EQ(1, left);
|
|
|
|
ASSERT_EQ(4, right);
|
|
|
|
// level 1, 1
|
|
|
|
GetNextLevelIndex(1, 1, -1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(1, left);
|
|
|
|
ASSERT_EQ(1, right);
|
|
|
|
GetNextLevelIndex(1, 1, 0, -1, &left, &right);
|
|
|
|
ASSERT_EQ(1, left);
|
|
|
|
ASSERT_EQ(1, right);
|
|
|
|
GetNextLevelIndex(1, 1, 1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(1, left);
|
|
|
|
ASSERT_EQ(3, right);
|
|
|
|
GetNextLevelIndex(1, 1, 1, 0, &left, &right);
|
|
|
|
ASSERT_EQ(4, left);
|
|
|
|
ASSERT_EQ(3, right);
|
|
|
|
GetNextLevelIndex(1, 1, 1, 1, &left, &right);
|
|
|
|
ASSERT_EQ(4, left);
|
|
|
|
ASSERT_EQ(4, right);
|
|
|
|
// level 1, 2
|
|
|
|
GetNextLevelIndex(1, 2, -1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(4, left);
|
|
|
|
ASSERT_EQ(3, right);
|
|
|
|
GetNextLevelIndex(1, 2, 0, -1, &left, &right);
|
|
|
|
ASSERT_EQ(4, left);
|
|
|
|
ASSERT_EQ(3, right);
|
|
|
|
GetNextLevelIndex(1, 2, 1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(4, left);
|
|
|
|
ASSERT_EQ(4, right);
|
|
|
|
GetNextLevelIndex(1, 2, 1, 0, &left, &right);
|
|
|
|
ASSERT_EQ(4, left);
|
|
|
|
ASSERT_EQ(4, right);
|
|
|
|
GetNextLevelIndex(1, 2, 1, 1, &left, &right);
|
|
|
|
ASSERT_EQ(4, left);
|
|
|
|
ASSERT_EQ(4, right);
|
|
|
|
// level 2, 0
|
|
|
|
GetNextLevelIndex(2, 0, -1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(0, left);
|
|
|
|
ASSERT_EQ(1, right);
|
|
|
|
GetNextLevelIndex(2, 0, 0, -1, &left, &right);
|
|
|
|
ASSERT_EQ(1, left);
|
|
|
|
ASSERT_EQ(1, right);
|
|
|
|
GetNextLevelIndex(2, 0, 1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(1, left);
|
|
|
|
ASSERT_EQ(1, right);
|
|
|
|
GetNextLevelIndex(2, 0, 1, 0, &left, &right);
|
|
|
|
ASSERT_EQ(1, left);
|
|
|
|
ASSERT_EQ(1, right);
|
|
|
|
GetNextLevelIndex(2, 0, 1, 1, &left, &right);
|
|
|
|
ASSERT_EQ(1, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
// level 2, 1
|
|
|
|
GetNextLevelIndex(2, 1, -1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(1, left);
|
|
|
|
ASSERT_EQ(1, right);
|
|
|
|
GetNextLevelIndex(2, 1, 0, -1, &left, &right);
|
|
|
|
ASSERT_EQ(1, left);
|
|
|
|
ASSERT_EQ(1, right);
|
|
|
|
GetNextLevelIndex(2, 1, 1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(1, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
GetNextLevelIndex(2, 1, 1, 0, &left, &right);
|
|
|
|
ASSERT_EQ(2, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
GetNextLevelIndex(2, 1, 1, 1, &left, &right);
|
|
|
|
ASSERT_EQ(2, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
// level 2, [2 - 4], no overlap
|
|
|
|
for (uint32_t f = 2; f <= 4; ++f) {
|
|
|
|
GetNextLevelIndex(2, f, -1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(f == 2 ? 2 : 3, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
GetNextLevelIndex(2, f, 0, -1, &left, &right);
|
|
|
|
ASSERT_EQ(3, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
GetNextLevelIndex(2, f, 1, -1, &left, &right);
|
|
|
|
ASSERT_EQ(3, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
GetNextLevelIndex(2, f, 1, 0, &left, &right);
|
|
|
|
ASSERT_EQ(3, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
GetNextLevelIndex(2, f, 1, 1, &left, &right);
|
|
|
|
ASSERT_EQ(3, left);
|
|
|
|
ASSERT_EQ(2, right);
|
|
|
|
}
|
2014-07-16 20:21:30 +02:00
|
|
|
delete indexer;
|
|
|
|
ClearFiles();
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
} // namespace rocksdb
|
|
|
|
|
|
|
|
int main(int argc, char** argv) {
|
2015-04-29 19:52:31 +02:00
|
|
|
rocksdb::port::InstallStackTraceHandler();
|
2015-03-17 22:08:00 +01:00
|
|
|
::testing::InitGoogleTest(&argc, argv);
|
|
|
|
return RUN_ALL_TESTS();
|
hints for narrowing down FindFile range and avoiding checking unrelevant L0 files
Summary:
The file tree structure in Version is prebuilt and the range of each file is known.
On the Get() code path, we do binary search in FindFile() by comparing
target key with each file's largest key and also check the range for each L0 file.
With some pre-calculated knowledge, each key comparision that has been done can serve
as a hint to narrow down further searches:
(1) If a key falls within a L0 file's range, we can safely skip the next
file if its range does not overlap with the current one.
(2) If a key falls within a file's range in level L0 - Ln-1, we should only
need to binary search in the next level for files that overlap with the current one.
(1) will be able to skip some files depending one the key distribution.
(2) can greatly reduce the range of binary search, especially for bottom
levels, given that one file most likely only overlaps with N files from
the level below (where N is max_bytes_for_level_multiplier). So on level
L, we will only look at ~N files instead of N^L files.
Some inital results: measured with 500M key DB, when write is light (10k/s = 1.2M/s), this
improves QPS ~7% on top of blocked bloom. When write is heavier (80k/s =
9.6M/s), it gives us ~13% improvement.
Test Plan: make all check
Reviewers: haobo, igor, dhruba, sdong, yhchiang
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D17205
2014-04-21 18:10:12 +02:00
|
|
|
}
|