Lynx/Galactica Net: a distributed, cache coherent multiprocessing system

The Lynx/Galactica Net system forms a clustered, shared memory multiprocessor with scalable performance. The Lynx component is a single-board multiprocessor using advanced, three-dimensional packaging technology to put four high performance RISC processors, memory, and I/O on a single 6m VME board. Galactica Net consists of a high speed mesh network combined with memory update hardware to maintain cache and memory coherence. Multiple Lynx computer nodes are connected together by Galactica Net to form shared memory systems with thousands of processors. A key feature of the Lynx/Galactica Net system is a tight coupling between the operating system and the coherency protocol for efficient, low-overhead coherency. An indication of the expected performance is provided through simulations using hardware-collected parallel address traces.<>