The Carbon research group is focused on research related to multicore architectures and software. We are a part of the Computer Science and Artificial Intelligence Lab at MIT.

Current Projects:

The Angstrom Project is investigating a fundamentally newants-tile computing architecture for 1000 core manycores to meet the challenges of exascale computing. Angstrom is based on two key ideas: a SElf-awarE Computational model called SEEC, and a fully distributed factored architecture for both hardware and software.

The SElf-awarE Computing (SEEC) model is a goal-oriented seec-overview computational model that radically increases developer productivity by abstracting traditional procedural programming into goals (e.g., “achieve the best possible chess move within 10s burning less than 20 W”) that are actuated in our self-aware, factored system.

fos

Factored Operating System (fos) is a new operating system targeting multicore, manycore, and cloud computing systems with scalability as the primary design constraint, where space sharing replaces time sharing to increase scalability.

Spatiotemporal Partitioning Strategies are a new set of design patterns strategyoverviewand accompanying selection methodology for developing parallel computations.  The patterns are based on the idea that programs can be partitioned by data or instructions and these partitionings can occur in time or space. This project presents the patterns and describes many case studies which demonstrate the selection methodology.

Graphite is a distributed parallel simulator for multicore architectures designed to simulate an application onGraphite: High Level Architecture 1000’s of cores by using dynamic binary translation on a given binary and uses hot-swappable modules for each part of the multicore chip.

atac_architecture_thumbnail

ATAC is a 1000-core manycore chip based on an All-to-All optical on-chip interconnect designed to improve bandwidth and programmability on future manycore chips.

smart_data_structures_overview

Smart Data Structures are a new class of parallel data structures that leverage online machine learning to self-tune themselves and help programmers achieve the performance potential of multicores without herculean effort.

The Application Heartbeatsheartbeats3 framework provides a simple, standardized way for applications to monitor their performance and express their goals, to make that information available to external observers, and to take actions to meet their goals. Heartbeats support the Self-Aware Computing (SEEC) model of computation.

Past Projects:raw

RAW is an early 16 core tiled multicore processor.

Alewife is a large-scale multiprocessor that integrates both cache-coherent, distributed shared memory and user-level message-passing in a single integrated hardware alewife16framework.