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 // For example, the sequence {20, 25, 41, 65432, 150000, 160000} would |
| 11 // be stored as: |
| 12 // A pair {20, 0} in |index_|. |
| 13 // 5, 16, 65391 in |deltas_|. |
| 14 // A pair {150000, 3} in |index_|. |
| 15 // 10000 in |deltas_|. |
| 16 // |index_.size()| will be 2, |deltas_.size()| will be 4. |
| 17 // |
| 18 // This structure is intended for storage of sparse uniform sets of |
| 19 // prefixes of a certain size. As of this writing, my safe-browsing |
| 20 // database contains: |
| 21 // 653132 add prefixes |
| 22 // 6446 are duplicates (from different chunks) |
| 23 // 24301 w/in 2^8 of the prior prefix |
| 24 // 622337 w/in 2^16 of the prior prefix |
| 25 // 47 further than 2^16 from the prior prefix |
| 26 // For this input, the memory usage is approximately 2 bytes per |
| 27 // prefix, a bit over 1.2M. The bloom filter used 25 bits per prefix, |
| 28 // a bit over 1.9M on this data. |
| 29 // |
| 30 // Experimenting with random selections of the above data, storage |
| 31 // size drops almost linearly as prefix count drops, until the index |
| 32 // overhead starts to become a problem a bit under 200k prefixes. The |
| 33 // memory footprint gets worse than storing the raw prefix data around |
| 34 // 75k prefixes. Fortunately, the actual memory footprint also falls. |
| 35 // If the prefix count increases the memory footprint should increase |
| 36 // approximately linearly. The worst-case would be 2^16 items all |
| 37 // 2^16 apart, which would need 512k (versus 256k to store the raw |
| 38 // data). |
| 39 // |
| 40 // TODO(shess): Write serialization code. Something like this should |
| 41 // work: |
| 42 // 4 byte magic number |
| 43 // 4 byte version number |
| 44 // 4 byte |index_.size()| |
| 45 // 4 byte |deltas_.size()| |
| 46 // n * 8 byte |&index_[0]..&index_[n]| |
| 47 // m * 2 byte |&deltas_[0]..&deltas_[m]| |
| 48 // 16 byte digest |
| 49 |
| 50 #ifndef CHROME_BROWSER_SAFE_BROWSING_PREFIX_SET_H_ |
| 51 #define CHROME_BROWSER_SAFE_BROWSING_PREFIX_SET_H_ |
| 52 #pragma once |
| 53 |
| 54 #include <vector> |
| 55 |
| 56 #include "chrome/browser/safe_browsing/safe_browsing_util.h" |
| 57 |
| 58 namespace safe_browsing { |
| 59 |
| 60 class PrefixSet { |
| 61 public: |
| 62 explicit PrefixSet(const std::vector<SBPrefix>& prefixes); |
| 63 |
| 64 // |true| if |prefix| was in |prefixes| passed to the constructor. |
| 65 bool Exists(SBPrefix prefix) const; |
| 66 |
| 67 private: |
| 68 // Maximum delta that can be encoded in a 16-bit unsigned. |
| 69 static const unsigned kMaxDelta = 256 * 256; |
| 70 |
| 71 // Maximum number of consecutive deltas to encode before generating |
| 72 // a new index entry. This helps keep the worst-case performance |
| 73 // for |Exists()| under control. |
| 74 static const size_t kMaxRun = 100; |
| 75 |
| 76 // Top-level index of prefix to offset in |deltas_|. Each pair |
| 77 // indicates a base prefix and where the deltas from that prefix |
| 78 // begin in |deltas_|. The deltas for a pair end at the next pair's |
| 79 // index into |deltas_|. |
| 80 std::vector<std::pair<SBPrefix,size_t> > index_; |
| 81 |
| 82 // Deltas which are added to the prefix in |index_| to generate |
| 83 // prefixes. Deltas are only valid between consecutive items from |
| 84 // |index_|, or the end of |deltas_| for the last |index_| pair. |
| 85 std::vector<uint16> deltas_; |
| 86 |
| 87 DISALLOW_COPY_AND_ASSIGN(PrefixSet); |
| 88 }; |
| 89 |
| 90 } // namespace safe_browsing |
| 91 |
| 92 #endif // CHROME_BROWSER_SAFE_BROWSING_PREFIX_SET_H_ |
OLD | NEW |