A "Measure of Transaction Processing" 20 Years Later
暂无分享,去创建一个
This article quantifies the price-performance improvements on two standard commercial benchmarks (DebitCredit and Sort) from 1985 to 2005. It shows that improvement has exceeded Moore’s law – largely due to (1) hardware improvements, (2) software improvements, (3) massive parallelism, and (4) changing from mainframe to commodity economics. Price-performance continues to improve faster than Moore’s law but per-processor and peak performance are improving more slowly. The sorting results in particular indicate little progress on per-processor speed in the last decade (probably due to slow improvement in memory bandwidth) and little overall progress since 2000 (when thousands of processors and disks were first used).
[1] Jim Gray,et al. Thousands of DebitCredit Transactions-Per-Second: Easy and Inexpensive , 2007, ArXiv.
[2] Robert Ramey. 2005 Performance / Price Sort and PennySort , 2005 .
[3] Michael Stonebraker,et al. A measure of transaction processing power , 1985 .
[4] Jim Gray,et al. Performance / Price Sort and PennySort , 1998 .
[5] Jim Wyllie,et al. Sorting on a Cluster Attached to a Storage-Area Network , 2005 .