High performance bulk memory system. Research report
We need a significant improvement in the technology of secondary memories. A bulk memory system with inexpensive memory doing fetches or writes in parallel (interleaving) provides one such technology. This interleaving is perfect for scientific computing in which one can usually design algorithms so that one need only access very long vectors in auxiliary storage. In these applications the effectiveness of the bulk memory is determined by the transfer rate. However, the bulk memory is even better suited for those problems, for example, as arise in commercial data processing, data base, and artificial intelligence applications, in which we wish to transfer small amounts of data. It is determined not by the transfer rate but by the latency and access time. In order to test the validity of our thesis, we have developed a bulk memory secondary storage device for a general-purpose supercomputer, the Floating Point Systems FPS-164. We present an overview of this system in Section 2. In Section 3, we describe the design philosophy of the interface to the FPS-164. To simplify the use of the bulk memory system, we have written a software system for the bulk memory which emulates a disk. This enables the user to use application software written to do I/O on a system with a peripheral disk virtually unchanged. We present the details of our software design in Section 4. The FPS-164 is a moderately priced, fast central processing unit with fast I/O ports. While array processors such as the FPS-164 have traditionally been used for engineering and scientific computations, we feel the lessons we learn with this system will carry over to the applications and other high performance computers.
- Research Organization:
- Yale Univ., New Haven, CT (USA). Dept. of Computer Science
- OSTI ID:
- 6712131
- Report Number(s):
- AD-A-140780/8; YALEU/DCS/RR-311
- Country of Publication:
- United States
- Language:
- English
Similar Records
Scientific Application Requirements for Leadership Computing at the Exascale
Automated Cache Performance Analysis And Optimization