Understanding Detailed Memory Performance Data

Lawrence Livermore National Laboratory Computer Science, 2015-16

Liaison(s): Todd Gamblin, Alfredo Gimenez, David Beckingsale
Advisor(s): Chris Stone
Students(s): Andrew Fishberg (PM), Paul Dapolito, Samuel Jackson, Xiaotian Wang

Mitos and MemAxes, a pair of memory analysis and visualization tools developed by Lawrence Livermore National Laboratory’s Scalability Team, aid supercomputer users in diagnosing high performance computing (HPC) speed bottlenecks. The LLNL Memory Clinic Team has expanded these tools to be more accessible and powerful for users. Specifically, the team has augmented widely used physics mesh simulation programs to provide detailed Mitos traces and explored machine learning techniques for automatically diagnosing performance problems.