Prefetching techniques for client server object-oriented database systems
Nils Knafla · ERA · 1999
4.18 Multithreading for CPU-intensive functions, like auditing. . . . .87 4.19 Effect of the number of prefetch threads at the database client. .88 5.1 Example of page dependencies . . .. . . . . . . . . . . . . . . . .91 5.2 Probability graph of object accesses . . . . . . . . . . . . . . . . .97 5.3 Savings of one prefetch dependent on the POD . . . . . . . . . . .101 5.4 Computation time to solve linear equations . . . . . . . . . . . . .102 5.5 Result of the simple simulation test .. . . . . . . . . . . . . . . . .105 5.6 Characteristics of the Demand applications under different cluster factors (elapsed times and number of demand page fetches). . .108 5.7 Characteristics of the Demand applications under different cluster factors (number of accessed pages and number of repeated page accesses ................................108 5.8 Characteristics of the Demand applications under different cluster factors (number of accessed objects per page) . . . . . . . . . . . .109 5.9 Result of the prefetch applications: P1, P1-DP, P2 and P2-DP. . .112 5.10 Result of the prefetch applications: P1-DP, P1-DP2, P2-DP and P2-DP2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... . .112 5.11 Total fetch time of all prefetch applications for transition probabilities from 1.0 to 0.5.........................113 5.12 Disk utilisation for cluster 90 applications ...............113 5.13 Performance of the Demand application and the three prefetch applications: P1, P2-DP, P2-DP2 with cluster factor 100 and cluster factor 90.................................114 5.14 Performance of the Demand application and the three prefetch applications: P1, P2-DP, P2-DP2 with cluster factor 80.......114 5.15 Improvements of the prefetch application P1 in % under the cluster factors of 90 and 80......................115 5.16 Improvements of the prefetch application P2-DP2 in % under the cluster factors of 90 and 80......................115 5.17 Normalised values for P1 considering prefetch accuracy, prefetch object distance and the number of prefetches for the applications with a cluster factor of 90 and 80....................116 5.18 Normalised values for P2-DP2 considering prefetch accuracy, prefetch object distance and the number of prefetches for the applications with a cluster factor of 90 and 80..............117 5.19 Prefetch application P1 with varied amount of client object processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .117 V 5.20 Prefetch application P2-DP2 with varied amount of client object processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .118 5.21 Effect of the buffer pool sizes of 10 and 30 frames on the Demand and P1 application with LRU replacement . . . . . . . . . . . . . .119 5.22 Effect of the buffer pool sizes of 50 and 100 frames on the Demand and P1 application with LRU replacement . . . . . . . . . .119 5.23 Effect of a decreasing number of buffer frames on the application with tp of 0.9 and a cluster factor of 90...............120 5.24 Improvement of the LRU-Prob replacement policy compared with a simple LRU policy for 10 and 30 buffer frames under a cluster factor of 80...............................120 5.25 Improvement of the LRU-Prob replacement policy compared with a simple LRU policy for 50 buffer frames under a cluster factor of8O..................................121 5.26 Effect of parallel disk accesses on the performance of the prefetch application P2-DP with n disks . . . . . . . . . . . . . . . . . . . .121 5.27 Reduction of fetch time of P2-DP over P2-DP2 ........... 122 6.1 Page server result .............................132 6.2 Object server result with threshold 0.0................133 6.3 Object server with all four thresholds . . . . . . . . . . . . . . . .134 6.4 Final object server result . . . . . . . . . . . . . . . . . . . . . . .