Memory Management

From HPCBugBase

Contents 1 Defect Type Description 2 Entries 3 Advice 4 Symptom 5 Other Findings and Contexts

[edit] Defect Type Description

This defect type is intended to cover defects related to various memory management issues. While memory management is a source of defects in any software domain, HPC applications seem to be more prone to this type of defects. There are several factors that contribute to this:

• In order to solve the largest problem possible, scientists often need to "push the limits" of available resources. Therefore, they are more likely to encounter memory problems than other type of applications that operate well below the safety limit.
• Since some defects of this type surface only in full-scale runs, it can be hard to reproduce when debugging.
• Parallelism makes this defect type harder to identify. If one of the processes fails due to a memory problem (e.g., due to out-of-memory) the entire program execution fails, but it is often hard to figure out what happened because failures may not seem to be linked to the actual cause.

[edit] Entries

Sub-types:

• Memory Allocation
• Memory Cleanup
• (Memory Hierarchy)

[edit] Advice

• Watch memory usage in each process carefully. Some platforms provide system support which allows users to keep track of memory usage during and/or after the program run.

[edit] Symptom

• Unexpected behavior when trying to scale up

[edit] Other Findings and Contexts