Clarete Crasta

Clarete Crasta

Senior Principal Engineer, HPE HPC/AI BU. Distributed Systems | HPC | Disaggregated Memory | FAM | Resiliency | Programming Models.

MODC: Resilience for Disaggregated Memory Architectures Using Task-Based Programming

Published in Workshop on Resource Disaggregation and Serverless (WORDS 2021) & arXiv:2109.05329, 2021

Presents MODC (Memory-Oriented Distributed Computing), a task-based programming framework for disaggregated memory architectures. Achieves resilience through task-level recovery and selective re-execution of failed tasks — drastically reducing mean time to recovery compared to whole-system checkpointing. Particularly relevant for long-running HPC workloads where failures occur after hours of computation.

Authors: K. Keeton, S. Singhal, H. Volos, Y. Zhang, R.C. Chaurasiya, C.R. Crasta, et al.

Venue: Workshop on Resource Disaggregation and Serverless (WORDS 2021); arXiv:2109.05329

Download paper

Recommended citation: K. Keeton, S. Singhal, H. Volos, Y. Zhang, R.C. Chaurasiya, C.R. Crasta, et al. (2021). "MODC: Resilience for Disaggregated Memory Architectures Using Task-Based Programming." Workshop on Resource Disaggregation and Serverless (WORDS 2021). arXiv:2109.05329.
Download Paper