Crash Dump Analysis Patterns (Part 176)

Sometimes, when an application is sluggish, periodically consumes CPU, it is possible to create a set of consecutive memory dumps of the same process to see the temporal development of any thread CPU consumption and figure out potential Spike Interval(s). For example, the following diagram was plotted from !runaway WinDbg command output for thread #1:

The 3rd and the 5th user process memory dumps in addition to increased CPU consumption also have corresponding non-waiting stack trace frames caught while executing some CPU instructions in ModuleA (not preempted with saved context). The first memory dump (yellow bar) with 437 ms user time spent out of 629 ms elapsed time also has a non-waiting stack trace but we consider it a normal application startup CPU consumption spike.

- Dmitry Vostokov @ + -

Leave a Reply

You must be logged in to post a comment.