Crash Dump Analysis Patterns (Part 211)

In process heap Memory Leak pattern we recommended acquiring sequential memory dumps spaced by 100MB. Unfortunately customers may send memory dumps spaced more closely, say by 10 - 20 MB or less after memory consumption growth already started some time in the past, for example, when they feel further process growth may impact their system performance. The analysis of process heap from memory dumps with enabled user mode stack database and corresponding UMDH log differences may show only Memory Fluctuation, where memory increases for specific stack trace allocations may follow by decreases or by small increases (Si is for memory dump size [horizontal bars], ti is for memory acquisition time):

In such cases it is difficult to choose among various local memory fluctuations to continue further investigation. However, a baseline process memory dump, for example, just after process start, helps to choose which stack trace allocations investigate first: those having bigger absolute memory allocation increase (Allocation Stack Trace B):

- Dmitry Vostokov @ + -

One Response to “Crash Dump Analysis Patterns (Part 211)”

  1. Dmitry Vostokov Says:

    We may need to assess object distribution for any anomalies:

Leave a Reply

You must be logged in to post a comment.