The development of the testbed system will require allocation of the hardware cost between the system components in such a way that bottlenecks to both high throughput and low response time are minimized. This requires the development of models of query workloads which allow storage, processing, network, and display requirements to be balanced.
The Alexandria project will be a user of the Computer Science Department's Meiko CS-2 parallel processor (NSF Grant No. CDA-9216202).
The Catalogue and Storage components will use dedicated processors built from workstation-class multiprocessors. Ingest will use dedicated PCs, as well as existing high-capacity digitizing stations. The Display components will run on PC and workstation class machines, such as are already part of the computing structure of the participating institutions.
The high-performance network technologies selected for the testbed will initially interconnect the CS-2, Catalogue, and Storage processors during the development phase. During deployment of Alexandria as a distributed system in the fourth year, the network will be updated to embrace whatever cost-effective technologies have become available by that time for distributing the system among the collaborating sites located around the country. The evolution of the NII will be a paramount consideration in the selection of the internetworking technology.
The storage system will consist of the high-end storage processor connected to three storage levels: a large buffer memory, secondary disk storage in the terabyte range built out of RAID disk arrays, and tertiary storage constructed from 8mm tape jukeboxes. SCSI disk storage is currently around $600/GB, 8mm tape storage at $55/GB. By the third year of the award, we anticipate these costs to improve by a factor of four.
The display technology will include standard resolution graphics on PC class machines and workstations/X-terminals. These are the predominant display devices that currently exist for users. High-resolution display devices, including prototype HDTV technology will also be used in the testbed. The displays devices will be driven by the Digital Library interface software running on high-end workstation processors.
Production versions of the testbed will likely leverage off of the commodity technologies being developed for the consumer/entertainment markets. Other research technologies that will be viable within ten years include virtual reality environments. The implementation of the testbed will steer toward these technologies.
Just as scalability of the testbed hardware towards new, higher-performing technologies is important, scalability towards lower, existing technologies is just as much of a concern. The ubiquitous PC is likely to be the most common display technologies. Many sites using a production version of Alexandria will use varyingly sophisticated hardware for ingest. 56Kb internet connections will dominate for years before being supplanted by the higher-speed NII and ATM-based network technologies.