|
Transcurrent Execution Model
Tree based indexing structures like Bit-Mapped indexes and Cubetrees have become essential for getting good performance in data warehousing, as they provide efficient multidimensional indexing for vast amounts of data. In this work we propose a new ``transcurrent execution model'' (TEM) for concurrent user queries against such indexes, which is based on the notion of detached non-blocking query patches. This is accomplished by processing index pages that are cached by the buffer manager and detaching execution of disk-resident patches of the query. Collaboration is achieved by synchronizing the detached patches and exploiting overlapping I/O among them. We further extend the TEM to support time constraints in the scheduling of the detached query patches and introduce the notion of data flow to achieve a steady flow of data pages to all queries. Our experiments demonstrate that the transcurrent query execution results in better utilization of the CPU and in high locality of I/O which in turn translates to substantial performance benefits in terms of query execution time, buffer hit ratio and disk throughput. These benefits increase as the workload in the warehouse increases and offer a highly scalable solution to the I/O problem of data warehouses. Transcurrent Papers
Transcurrent People |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||
