P. Larson and G. Graefe. Memory management during run generation in external sorting. Proc. SIGMOD 1998, pages 472--483, 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....arising from this discussion is how long the runs can be if we want to overlap I O and computation. Knuth [15, Section 5.4.1] describes an algorithm that achieves average run length 2M . A recent implementation that even works for variable length records has been described by Larson and Graefe [16]. However, this algorithm is not cache e#cient and requires an additional pointer for each element in the input. We therefore outline a relatively simple reformulation that is space e#cient even for small records, cache e#cient, and provably allows overlapping of I O and computation. A more ....

P. Larson and G. Graefe. Memory management during run generation in external memory. In SIGMOD, pages 472--484. ACM, 1998.

....similar to a sensor network [3, 49] Kabra and DeWitt proposed to reoptimize parts of queries after blocking operators [24] There is also a lot of work on adaptive query operators, an area we believe to be relevant to sensor networks. Examples include work on memory adaptive sorting and hashing [13, 28, 30, 34, 53, 54], and online aggregation algorithms [15, 18, 39, 48] Eddies push the idea of feedback on a tuple by tuple basis in online aggregation to adapting join orders at the same frequency [4] Other relevant work includes sequence query processing [42, 43] and temporal and time series databases [52] ....

P.-. Larson and G. Graefe. Memory management during run generation in external soting. In Haas and Tiwary [14], pages 472-483.

.... fewer runs must be balanced with the different I O pattern and the disadvantage of more complex memory management [3] 0,0 0,4 0,8 1,2 1,6 2,0 2,4 2,8 3,2 3,6 4,0 0 2000000 4000000 6000000 8000000 10000000 12000000 Number of Tuple Merge Sort TempTris Figure 6 2:Ratio merge sort and TempTris In [11] it is shown that it is possible to create 1.8 times larger runs than the workspace. But this has no influence on the measurement. As we set the available cache M used by the merge sort and TempTris algorithm at most half of the size of the measured sets and smaller than m M , external ....

P. Larson and G. Graefe, "Memory Management During Run Generation in External Sorting," presented at SIGMOD,International Conference on Management of Data, Seattle, Washington, USA, 1998.

....and Section 7 discusses future directions. 2 Related Work Much of the related research has focused on improving the query execution time, mainly by minimizing the stalls due to memory hierarchy when executing an isolated task. There are a variety of algorithms for fast sorting techniques [1][12][15] that propose optimal data placement into memory and sorting algorithms that minimize cache misses and overlap memory related delays. In addition, several cache conscious techniques such as blocking, data partitioning, loop fusion, and data clustering were evaluated [17] and found to improve ....

....DBMS performance should focus on minimizing all three kinds of stalls to effectively decrease the execution time. 5. 2 Memory stalls In order to optimize performance, a major target of database research has been to minimize the stall time due to memory hierarchy and disk I O latencies [1][12][15] 17] Several techniques for cache conscious data placement have been proposed [3] to reduce cache misses and miss penalties. Although these techniques are successful within the context in which they were proposed, a closer look at the execution time breakdown shows that there is significant ....

P. . Larson, and G. Graefe. Memory management during run generation in external sorting. In Proceedings of the 1998 ACM SIGMOD Conference, June 1998.

....future directions are contained in Section 6. 2 Related work Much of the database research has focused on improving the query execution time, mainly by minimizing the stalls due to memory hierarchy when executing an isolated task. There are a variety of algorithms for fast sorting techniques [1][12][15] that propose optimal data placement into memory and sorting algorithms that minimize cache misses and overlap memory related delays. In addition, several cache conscious techniques such as blocking, data partitioning, loop fusion, and data clustering were evaluated [17] and found to improve ....

....should focus on minimizing all three kinds of stalls to effectively decrease the execution time. 5. 2 Memory stalls In order to optimize performance, the main target of database researchers in the past has been to minimize the stall time due to latencies caused by the memory hierarchy [1][12][15] 17] Several techniques for Figure 5.2: Contributions of the five memory components to the memory stall time Indexed range selection 0 20 40 60 80 100 B C D L1 D stalls L1 I stalls L2 D stalls L2 I stalls ITLB stalls Join 0 20 40 60 80 100 A B C D Sequential range se lection ....

P. Å. Larson, and G. Graefe. Memory management during run generation in external sorting. In Proceedings of the 1998 ACM SIGMOD Conference, June 1998.

No context found.

P. Larson and G. Graefe. Memory management during run generation in external sorting. Proc. SIGMOD 1998, pages 472--483, 1998.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC