🏛️ Supercomputing History: MHPCC and the Mid-90s Top Tier

📌 Context and Historical Standing

In 1994 and 1995, the Maui High Performance Computing Center (MHPCC) was globally recognized as one of the elite centers for high-performance computing. Officially launched in late 1994 with a 400-node unclassified IBM SP2 production system, it later operated 512 SP2 nodes in aggregate across separate production, classified, and training clusters.

At the time, MHPCC was billed as the 6th largest unclassified technical computing center in the world.

🥇 The Global Top Tier, 1994-1995

The unclassified facilities and systems ranking ahead of or alongside MHPCC on the official TOP500 lists included:

1. Sandia National Laboratories (U.S.): Maintained the #1 spot in 1994 using the massive Intel Paragon XP/S 140 cluster.
2. National Aerospace Laboratory (Japan): Operated the Numerical Wind Tunnel, a custom vector parallel Fujitsu system that dominated global benchmarks.
3. Los Alamos National Laboratory (U.S.): Powered by a massive 1,056-core Thinking Machines CM-5 and subsequent Cray T3D systems.
4. Oak Ridge National Laboratory (U.S.): Anchored by the Intel Paragon XP/S 150, holding a top-three global position by early 1995.
5. The Cornell Theory Center (U.S.): A premier academic peer that deployed its own production 512-node IBM SP2 system in early 1995.
6. National Center for Supercomputing Applications (NCSA) (U.S.): Based at UIUC, running a highly ranked Thinking Machines CM-5/512.
7. Minnesota Supercomputer Center (U.S.): An academic powerhouse operating a top-tier CM-5 cluster.
8. National Security Agency (NSA) (U.S.): Maintained equivalent hardware, including massive CM-5 units, though a large portion of their aggregate center performance remained classified.

🛠️ System Architecture: The IBM SP2

The backbone of MHPCC's mid-90s dominance was the IBM SP2, a milestone architecture that proved commercial RISC-based parallel computing could outpace traditional monolithic supercomputers.

📐 Structural Overview

• System Type: Massively Parallel Processing (MPP) Multicomputer.
• Topology: Distributed Memory / Shared-Nothing Architecture. Each node ran its own private memory and its own individual instance of the AIX operating system, IBM's Unix.
• Programming Model: Explicit Message Passing. Applications relied heavily on Message Passing Interface (MPI) and IBM's Message Passing Library (MPL) to coordinate workloads across processors.

Source: converted from the Obsidian note "MHPCC History" and published as context for the preserved 1994 MHPCC website archive.