🌐 MHPCC Network Architecture, circa 1994

WAN And Internet Presence

In 1994, the Maui High Performance Computing Center connected to the global internet backbone through a dedicated T1 circuit operating at 1.544 Mbps. From Kihei, Maui, traffic crossed inter-island infrastructure, transpacific undersea cable paths, and West Coast gateways before reaching mainland backbone networks.

Physical Routing Path

• Kihei, Maui: Outbound supercomputer traffic exited the MHPCC facility.
• Oahu transit: Traffic hopped inter-island to data hubs on Oahu.
• Transpacific cable path: Data traveled by undersea cable toward the mainland.
• Los Angeles, California: The transpacific link landed at the West Coast gateway.

🏛 Routing And Backbone Context

• Autonomous System Number: MHPCC was registered as AS3381, also known as MHPCC-AS, on January 7, 1994.
• Primary backbone provider: The wide-area path was associated with ANSNET, Advanced Networks and Services.
• Upstream context: The center connected into the NSFNET-era backbone environment and early DoD High Performance Computing networking efforts.

⚠️ Propagation Delay And TCP Bottlenecks

Running a major IBM SP2 installation from Maui over a single T1 link exposed wide-area networking limits that were easy to overlook on the mainland. The distance to Los Angeles made MHPCC a practical example of the early Long Fat Network problem.

• High propagation latency: The long physical path added substantial delay.
• TCP window limits: Early-1990s TCP behavior could stall over long-distance, high-delay paths.
• Throughput choking: Latency could be misread as congestion.
• Pipe underutilization: The T1 could fail to reach its full useful capacity.

🖥 Early Web Space

• Launch date: The official MHPCC World Wide Web server launched in August 1994.
• Web scale: This was early enough that the public web still had only a few thousand active websites.
• Browser context: Many users reached the web through NCSA Mosaic.
• Domain email: External communication used the help@mail.mhpcc.edu gateway shown in the original archive.

🔌 Internal Kihei Facility Network

The Kihei datacenter had to connect a heterogeneous computing environment, not a single monolithic machine. MHPCC's public-facing production system was the 400-node IBM 9076 SP2. By the end of the buildout, the center operated 512 SP2 nodes in aggregate across production, classified, and training clusters.

IBM SP2 High-Performance Switch

• Role: Primary node-to-node interconnect inside the SP2 frames.
• Technology: IBM's proprietary High-Performance Switch, originally code-named Vulcan.
• Topology: Multi-stage bidirectional switching fabric using crossbar switch boards.
• Performance context: Low-latency message passing between AIX nodes, with roughly 40 MB/s, or 320 Mbps, per link.

HIPPI Fabric

• Speed: 800 Mbps parallel transmission over heavy multi-pair copper cabling.
• Role: High-speed data path between SP2 systems, large storage arrays, and visualization systems.
• Use case: Moving large scientific and imaging data sets to Silicon Graphics visualization workstations and storage resources.

FDDI Backbone

• Speed: 100 Mbps dual-ring token-passing fiber network.
• Role: Campus and facility backbone.
• Connected systems: Sun workstations, secondary servers, administrative systems, X-terminals, and supercomputing room infrastructure.

🎛 Wide Area Frame Relay

In addition to the dedicated T1 backbone to Los Angeles, MHPCC used Frame Relay as a secondary wide-area data link technology.

• Burst economics: Frame Relay allowed bursty traffic over shared infrastructure with a lower committed information rate.
• Local and partner access: It was useful for connecting Hawaii institutions, commercial R&D partners, and state agencies.
• Experimental programs: Frame Relay and ISDN-era networking were part of the broader context for early remote-access, telemedicine, and disaster-relief communication pilots.

🖥 Multi-Vendor Systems Connected By The Network

• IBM processing core: 400-node unclassified IBM 9076 SP2, plus separate classified and training SP2 clusters.
• SGI visualization tier: Silicon Graphics systems used for rendering scientific, satellite, and atmospheric data.
• Sun infrastructure: Sun SPARC workstations and servers used for development, compilation, monitoring, and researcher access.

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