NuMesh

NuMesh: An architecture optimized for scheduled communication. The NuMesh group is one of the three subgroups of the MIT Laboratory for Computer Science’s Computer Architecture Group. The group has one full-time and one part-time faculty member, five graduate students, and one member of research staff. NuMesh is a packaging and interconnect technology supporting high-bandwidth systolic communications on a 3D nearest-neighbor lattice; our goal is to combine Lego-like modularity with supercomputer performance. To date, the primary focus of the project has been the class of applications whose static communication patterns can be precompiled into independent and carefully choreographed finite state machines running on each node. Several extensions of the NuMesh to more general communication paradigms have been implemented. Our primary hardware platform is composed of 6”x6” 40MHz boards running locally-designed finite-state machine hardware, with 32 bit busses connecting each of its four neighbors in a two-dimensional mesh, and a 1024-element bi-directional FIFO to an optional CPU board. We have designed CPU boards incorporating a SPARC processor with 8M of RAM; a TMS 320C30 DSP processor; and other boards, including video hardware and the Sun host interface. Another hardware platform is 1”x1” boards with integral 6811 microprocessors and four eight-bit connections to neighboring nodes. Rather than the two-dimensional interconnect currently used in our primary hardware platform, these boards connect together in a four-neighbor diamond lattice configuration. We have a 100-node configuration available for experimenting with diamond-lattice routing and algorithms, as well as heat-dissipation tests using the nodes’ built-in heating element and thermometer. Software work is ongoing to examine issues involved in precomputed routing to optimally utilize the capabilities of the hardware.

Keywords for this software

Anything in here will be replaced on browsers that support the canvas element