Igor Ljubuncic

Be Kind, Rewind: Checkpoint a Restore Capability for Improving Reliability of Large-Scale Semiconductor Design

Intels chip design run in a large-scale globally distributed environment with 600,000 cores. In the current semiconductor market scenario, a combination of factors such as time to market pressure, explosive growth in the mobile market segment and upcoming new markets has led to a significant increase in the demand for and reliability of computing resources. Checkpointing is a capability that can make a significant improvement in improving reliability, however, there is no mature solution that allows periodic snapshots of running compute jobs for replay them at a later time in a consistent manner in a large scale environment. Intel IT has partnered with the Northeastern University (NEU) Distributed Multi-Threaded Checkpointing (DMTCP) team to improve their checkpoint & restore solution for the design computing environment. This paper elaborates on the innovative technological breakthroughs, industry-academy partnership as well as the open-source contribution.