EPTRACE Manual

EPTRACE Manual—528811-002
1-1
1 Introducing EPTRACE
EPTRACE is a performance investigation tool designed for the Guardian and OSS
environments. It reports detailed counts of performance related trap events and
provides a trace log of these events so you can locate, diagnose, and resolve trap-
causing locations in a program.
EPTRACE reports two types of information in the trace log:
•
Counters
Reports individual counters for each process, including interrupt processes.
The per-CPU totals equal the sum of all process counters.
•
Traces
•
Locates particular events when you enable tracing. To report traces you must
enable the tracing function using the START command
•
EPTRACE reports these event trace items:
•
Current program name and PIN
•
Time
•
Event Type
•
Stumble subclass
•
Trapped program location (native IA-64 IP, TNS P and/or translated IA-64I
IP. Locations within system libraries are included, but not separately
counted or separately traced.
•
Trapped operand addresses
Performance Considerations
•
Unless there are a significant number of events, EPTRACE should not impact
application performance
•
You may run only one EPTRACE session on a CPU at any given time.
•
EPTRACE can consume a significant amount of process time and resident
memory when there is a high incidence rate of traced events in a program.
•
EPTRACE does not trace or count atomic operations, expected mode transitions,
RP wraparounds, TLB misses, or disabled flow traps.
•
The trace buffer is shared between all traced processes on a specific CPU.