Chromium Code Reviews| OLD | NEW |
|---|---|
| (Empty) | |
| 1 // Copyright (c) 2011 The Chromium Authors. All rights reserved. | |
| 2 // Use of this source code is governed by a BSD-style license that can be | |
| 3 // found in the LICENSE file. | |
| 4 // | |
| 5 // A read-only set implementation for |SBPrefix| items. Prefixes are | |
| 6 // sorted and stored as 16-bit deltas from the previous prefix. An | |
| 7 // index structure provides quick random access, and also handles | |
| 8 // cases where 16 bits cannot encode a delta. | |
| 9 // | |
| 10 // This structure is intended for storage of sparse uniform sets of | |
| 11 // prefixes of a certain size. As of this writing, my safe-browsing | |
| 12 // database contains: | |
| 13 // 653132 add prefixes | |
| 14 // 6446 are duplicates (from different chunks) | |
| 15 // 24301 w/in 2^8 of the prior prefix | |
| 16 // 622337 w/in 2^16 of the prior prefix | |
| 17 // 47 further than 2^16 from the prior prefix | |
| 18 // For this input, the memory usage is approximately 2 bytes per | |
| 19 // prefix, a bit over 1.2M. The bloom filter used 25 bits per prefix, | |
| 20 // a bit over 1.9M on this data. | |
| 21 // | |
| 22 // Experimenting with random selections of the above data, storage | |
| 23 // size drops almost linearly as prefix count drops, until the index | |
| 24 // overhead starts to become a problem a bit under 200k prefixes. The | |
| 25 // memory footprint gets worse than storing the raw prefix data around | |
| 26 // 75k prefixes. Fortunately, the actual memory footprint also falls | |
| 27 // continuously. If the prefix count increases the memory footprint | |
| 28 // should increase approximately linearly. | |
| 29 // | |
| 30 // TODO(shess): Write serialization code. Something like this should | |
| 31 // work: | |
| 32 // 4 byte magic number | |
| 33 // 4 byte version number | |
| 34 // 4 byte |index_.size()| | |
| 35 // 4 byte |deltas_.size()| | |
| 36 // n * 8 byte |&index_[0]..&index_[n]| | |
| 37 // m * 2 byte |&deltas_[0]..&deltas_[m]| | |
| 38 // 16 byte digest | |
| 39 | |
| 40 #ifndef CHROME_BROWSER_SAFE_BROWSING_PREFIX_SET_H_ | |
| 41 #define CHROME_BROWSER_SAFE_BROWSING_PREFIX_SET_H_ | |
| 42 #pragma once | |
| 43 | |
| 44 #include <vector> | |
| 45 | |
| 46 #include "chrome/browser/safe_browsing/safe_browsing_util.h" | |
| 47 | |
| 48 namespace safe_browsing { | |
| 49 | |
| 50 class PrefixSet { | |
| 51 public: | |
| 52 explicit PrefixSet(const std::vector<SBPrefix>& prefixes); | |
| 53 | |
| 54 // |true| if |prefix| was in |prefixes| passed to the constructor. | |
| 55 bool Exists(SBPrefix prefix) const; | |
| 56 | |
| 57 private: | |
| 58 // Maximum delta that can be encoded in a 16-bit unsigned. | |
| 59 static const unsigned kMaxDelta = 256 * 256; | |
| 60 | |
| 61 // Maximum number of consecutive deltas to encode before generating | |
| 62 // a new index entry. This helps keep the worst-case performance | |
| 63 // for |Exists()| under control. | |
| 64 static const size_t kMaxRun = 100; | |
| 65 | |
| 66 // Top-level index of prefix to offset in |deltas_|. Each pair | |
|
lzheng
2011/02/05 01:03:42
You might want to give an example. It is a little
Scott Hess - ex-Googler
2011/02/07 19:20:23
Adding a comment in the file header.
| |
| 67 // indicates a base prefix and where the deltas from that prefix | |
| 68 // begin in |deltas_|. The deltas for a pair end at the next pair's | |
| 69 // index into |deltas_|. | |
| 70 std::vector<std::pair<SBPrefix,size_t> > index_; | |
| 71 | |
| 72 // Deltas which are added to the prefix in |index_| to generate | |
| 73 // prefixes. Deltas are only valid between consecutive items from | |
| 74 // |index_|, or the end of |deltas_| for the last |index_| pair. | |
| 75 std::vector<uint16> deltas_; | |
| 76 | |
| 77 DISALLOW_COPY_AND_ASSIGN(PrefixSet); | |
| 78 }; | |
| 79 | |
| 80 } // namespace safe_browsing | |
| 81 | |
| 82 #endif // CHROME_BROWSER_SAFE_BROWSING_PREFIX_SET_H_ | |
| OLD | NEW |