John Glenn Thompson

An architecture for storage-hosted application extensions

The very rapid growth of data-intensive computing makes it attractive to perform computations locally, where data is stored. Large storage systems based on standard system technologies with server virtualization capabilities make it feasible to deploy application-specific processing onto the storage system, without jeopardizing the availability of the core storage service or degrading performance. Moreover, price and capacity differences between mainframes and these storage systems make this deployment attractive. We describe the design of a prototype system by which the IBM DS8000™ storage system can host application extensions, called adjuncts, that improve the operation of IBM z/OS® (mainframe) applications. These extensions process large amounts of data in operations such as searching, sorting, and indexing so that the host application need not even access most of the data. The benefits of application extensions result from applying system resources more efficiently. Application processing at the storage system magnifies the total throughput that can be achieved by the host application. Furthermore, by avoiding the transmission of large volumes of data through multiple hardware and software layers, processing often takes a shorter time at a lower cost.