Desktop batch processing

Today, online transaction processing (OLTP) applications can downsize from mainframes to microprocessors. Commodity database systems, operating systems, and hardware came of age in 1993/spl minus/they surpassed the online transaction processing performance of proprietary solutions. There are lingering doubts about downsizing batch transaction processing applications. The doubts center on the ability of microprocessor hardware to handle the high I/O bandwidth required by batch processing, and on doubts that microprocessor systems offer the software services and utilities key to batch processing applications. This paper reviews the impressive progress of made by commodity software and hardware in processing OLTP workloads. The discussion is quantitative because the Transaction Processing Performance Council defined a set of benchmarks that characterize OLTP and that quantify price and performance. Discussion then turns to batch transaction processing. There is less consensus on the characteristics of batch transaction processing. Consequently, much of the discussion focuses on requirements. The discussion ends with some performance measurements of utilities running on DEC Alpha AXP microprocessors and on commodity disks. These results indicate that microprocessors today have the capacity to process batch workloads at mainframe speeds. We predict that over the next few years, batch-processing software exploiting parallel processing will emerge. This, combined with commodity hardware, will provide both superior performance and price/performance ratio.<>