Aspects of Metrics
Within an individual machine:
BetwMetrics are gathered from major components of a system:
|
|
System Metrics
Both Mercury Deep Diagnostics 
and Mercury Profiler
monitor these metrics:
- CPU: Percentage utilization of the CPU. Note that the current release of Deep Diagnostics
supports only a single CPU.
- Memory: The percentage of memory in use.
- Virtual Memory: The percent of virtual memory in use.
- Context Switches per Second:
How many times per second the operating system switches threads.
- Page Ins per Second:
The number of operating system pages being brought into the system per second.
- Page Outs per Second:
The number of operating system pages being swapped out of the system per second.
- Swap Ins per Second:
The number of pages read from swap space per second. (This does not apply to Windows systems.)
- Swap Outs per Second:
The number of pages written to swap space per second. (This does not apply to Windows systems.)
- Disk Bytes per Second:
The number of bytes being transferred to/from the disk per second.
- Disk I/O per Second:
The total number of transfer operations to and from the disk per second.
Additionally, Deep Diagnostics measures:
- Network Throughput:
The number of bytes transmitted per second between any two
systems running Deep Diagnostics Agents.
- Network Bytes per Second:
The number of bytes being transferred to/from the network per second.
- Network I/O per Second:
The total number of transfer operations to and from the network per second.
- Network Latency:
The elapsed time of a single call between two systems running Deep Diagnostics Agents (in milliseconds).
- BEAJMX: Allows the monitoring of BEA WebLogic (6.1 or later) via the Java Management
eXtensions (JMX). To collect these metrics, you must installed the support for
JMX on the System Under Test.
- CPU Time —
the time that a given method invocation spends
actually executing on the processor,
including the time spent on the CPU for child calls.
- Exclusive CPU Time excludes time spent on the CPU for child calls.
These are available only after the property use.cpu.timestamps in the
capture.properties file within the probe's /etc directory is set to "true"
(from "false" by default) to capture per-thread CPU timings.
- Latency
is the wall-clock time from the time of the method start
to the time of the method finish, for all methods.
- Avg. Latency is the mean Latency calculated by dividing total Latency by the number of calls.
- Throughput
- Allocation
- Deallocation
- Live Object
|
|
|
|
IBM WebSphere PMI
Metrics
