uhler A Systems Concepts PDP-10 Mar83

This document is an interoffice memorandum from Mike Uhler to Peter Hurley and Jan Jaferian, dated January 5, 1983 (with follow-up memos on Jan 17, 1983), concerning the requirements for a PDP-10 processor designed by Systems Concepts.

Key Points:

1. Core Challenge & Recommendation: The primary challenge is the lack of a complete, formal definition of the PDP-10 architecture. The memo strongly recommends using the "Jupiter" design as the architectural baseline for Systems Concepts, rather than the older KL10, due to Jupiter's bug fixes, extended addressing, and other improvements.
2. Compatibility Goals:
   • Non-privileged user-mode programs should detect no differences when run on a Jupiter versus the Systems Concepts design.
   • Ideally, exec-mode programs should also be fully compatible to avoid past issues with hardware-software coupling.
3. Architectural Definition Sources: The formal definition of the PDP-10 Architecture for this project is comprised of:
   • The Processor Reference Manual (annotated with Digital's latest approved updates and Mike Uhler's corrections).
   • Mike Uhler's memo on "Extended Addressing."
   • "KCIO.MEM," which defines Jupiter's exec-mode instructions and processor-dependent information, with mandatory adherence.
4. Verification and Testing:
   • A comprehensive set of diagnostic and verification programs are specified, including Jupiter CPU diagnostics, extended functionality tests, and programs for debugging Jupiter microcode/hardware (e.g., DCKxx series, EAT, TPAGER). These programs use Jupiter-specific instructions and console protocols and must run without modification on the Systems Concepts machine.
   • Performance verification involves running specific COBOL, FORTRAN, and miscellaneous benchmark programs (e.g., COMPUT, SPEED, SANDIA Tests) in standalone mode. The Systems Concepts CPU's aggregate performance must be equal to or better than a KL10E system, with no individual instruction timing exceeding 1.2 times the KL10E, and floating-point/byte/string operations being at least as fast.
   • The ultimate measure of correct implementation is a fully running monitor and user-mode software.
5. Areas for Further Consideration: The memo outlines several crucial topics for any future PDP-10 design, emphasizing that they are not yet well-defined and require further analysis:
   • Performance Measurement: Developing methodologies to accurately measure machine performance, possibly using opcode histogram data and instruction timing comparisons.
   • Extended Addressing: Understanding and testing the performance implications of increased extended addressing use.
   • Translation Buffer & Cache: Optimizing their organization and size (e.g., considering multi-way associative translation buffers) based on Jupiter's experience.
   • Performance Instrumentation: Incorporating hardware and microcode instrumentation points to detect bottlenecks.
   • I/O Structure: Ensuring compatibility with new I/O architectures (CI/NI interfaces) and Jupiter's console protocol.
   • RAMP (Reliability, Availability, Maintainability, Performance): Defining error detection, recovery, and monitor handling of exceptions.
   • Microcode Space: Ensuring sufficient unused microcode space (a minimum of 10%) for future bug fixes and enhancements.
   • Problem Resolution: Establishing a formal process for resolving architectural and implementation issues and approving design changes.
6. Acceptance Criteria: Acceptance will be granted when all specified programs run correctly, Systems Concepts' diagnostics are verified, user-mode compatibility is confirmed by Digital, design modification responsibilities are assigned, and Digital provides an evaluation report. This process is expected to be completed within six months of the unit's delivery. Successful demonstration of remote serviceability is also required.