1 / 9

Topics 10: Cache Conscious Indexes

Topics 10: Cache Conscious Indexes. As main memory gets cheaper, it becomes affordable to build computers with large memories. In future databases all data but few large tables will be memory-resident. Therefore is it important to build efficient main-memory indexes.

meris
Download Presentation

Topics 10: Cache Conscious Indexes

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Topics 10: Cache Conscious Indexes • As main memory gets cheaper, it becomes affordable to build computers with large memories. • In future databases all data but few large tables will be memory-resident. • Therefore is it important to build efficient main-memory indexes. • These indexes should consider the hierarchical memories and the memory-access bottleneck. Advanced Database Technologies

  2. Characteristics of cache conscious indexes • They should cluster data according to the access pattern; data that are likely to be accessed together (or in sequence) should be close in memory. • They should compress information, so that only useful data are fetched in cache. This means that only comparison keys and reference pointers to searched data should be in the index. • They should not be much larger than the indexed information. Advanced Database Technologies

  3. Why is binary search poor? • If the searched array is large, the number of cache misses is determined by the search comparisons: O(log2n). • This is because from the information fetched in the cache, only one search key will be used. cache-line (128 bytes) Cache: MMem: ... 539 545 568 579 582 589 595 602 609 612 617 623 625 ... current key comparison Advanced Database Technologies

  4. Enhanced Main Memory B+-trees • Although the B+-tree is a secondary memory index, it can be used for search in main memory. • The node size of the tree is set to a multiple of the cache linesize (e.g., 1 node=2 cachelines). • Now the number of cache misses equals the number of tree nodes accessed at search: O(logFn), where F is the fanout of the tree. Advanced Database Technologies

  5. Problems of Main Memory B+-trees • Nodes contain as many pointers as key values. • Many key values can be compared in a node during search. On the other hand, only one pointer will be followed. • Binary search in a node could be expensive (requiring many comparisons) Advanced Database Technologies

  6. 12 19 24 31 The Cache Sensitive Search (CSS) tree • Same as B+-tree, but does not store pointers. • The children of each node are stored sequentially, thus pointers are induced by positional memory offsets. CSS-tree B+-tree 24 12 19 31 4 8 9 12 13 17 19 21 23 24 27 29 31 34 38 4 8 9 12 13 17 19 21 23 24 27 29 31 34 38 Advanced Database Technologies

  7. The Cache Sensitive Search (CSS) tree (cont’d) • The CSS tree is suitable only for static data. • The capacity of each node is double the capacity of an B+-tree node. • Thus the height (and search cost) of the tree is reduced. • Another trick used by the CSS tree is hard-coding binary search by if-else statements. Advanced Database Technologies

  8. Hard-coding binary search • Normal binary search • Binsearch(key,C,start,end)= Binsearch(key,C,mid,end) if key>C[mid] Binsearch(key,C,start,mid) if key<C[mid] Follow C[mid] if key=C[mid]. • Augmented binary search • if (key<C[mid]) then • if (key<C[mid/2]) then ... • else if (key>C[mid/2]) then ... • else follow C[mid/2] • else if (key>C[mid]) then • if (key<C[3mid/2]) then ... • else if (key>C[3mid/2]) then ... • else follow C[3mid/2] • else follow C[mid] Advanced Database Technologies

  9. Presentation material • A dynamic version of the CSS-tree: the cache conscious B+-tree • An improved version of the cache conscious B+-tree (optional reading) • Cache conscious R-trees Advanced Database Technologies

More Related