|
Problem(Abstract) |
Ballpark performance degradation metrics when using PMI,
JVMPI and PMI request metrics with ARM. |
|
|
|
Resolving the
problem |
PMI
overhead is 3.0% when using the standard level setting, where most of the
counters are enabled except for those with maximum level, and when running
Tivoli® Performance Viewer at the same time. When JVMPI
counters are enabled, significant performance degradation is observed.
JVMPI
has 4 types of data: GC, Thread, Object, and Monitor. In a full JVMPI
configuration where all 4 are enabled, the impact from the full JVMPI
data is about 55%.
For PMI request
metrics with ARM
enabled, and no filters (instrumenting all requests), the overhead in
WebSphere® Application Server 5.1.x is in approximately 7%. This overhead
can be further mitigated by using the filtering capabilities.
If both PMI
and PMI request
metrics are enabled, the expected added performance degradation is
10%. |
|
|
|
|
Cross Reference information |
Segment |
Product |
Component |
Platform |
Version |
Edition |
Application Servers |
Runtimes for Java Technology |
Java SDK |
|
|
|
|
|
|