This page lists performance metrics, i.e. properties of processes that we may wish to measure. It is intended to make the definitions of these properties clear so that different agents can use the data reliably.
See also the PLANETS property definitions, which provide a similar starting point:
| Name | Description | Data Type | Units | Identifier | Comment |
|---|---|---|---|---|---|
| Wall clock time | The elapsed time, as would be measured by a clock on the wall. | xsd:double | seconds | planets:service/exec/wallclock action:/performanceEfficiency/timeBehaviour/elapsedTime |
(see here ) see D14.1 (aligned with SQUARE quality attributes) |
| cpu time | Processing time used by the CPU, measured in milliseconds. | xsd:integer | millisecond | action:/performanceEfficiency/timeBehaviour/cpuTime | one could define this as the cpu time used by the process, or as the cpu time used by the process and any children (doable on *nix with top, for example). if this is defined in the first way, we would need another attribute for the second interpretation since this is more relevant. |
| peak memory |
Peak memory used during processing |
xsd:double |
MB |
action:/performanceEfficiency/resourceUtilization/peakMemory |
for Java and comparable processes, one could distinguish between measures outside the VM and inside the VM, but it is really easier to see this as the same attribute, just a different measurement method. This counts mosrly for memory, since JVM startup time is constant, but reserved memory is dependent on the call parameters |
| average memory |
Average memory used during processing |
xsd:double |
MB |
action:/performanceEfficiency/resourceUtilization/averageMemory |
|
| used memory share |
Percentage of reserved memory actually used |
xsd:double |
percentage |
action:/performanceEfficiency/resourceUtilization/memoryShare |
(mostly relevant for Java processes, if at all) |
| average load |
Processing load on machine during execution |
xsd:double |
percentage |
this isnt a process measure, of course, but an interesting context measure that for example can be delivered using top |
Labels:
None
2 Comments
comments.show.hideJan 27, 2012
Christoph Becker
We are adding some metrics. The assumption is that we are referring to ONE process running ONE object, so that with knowing the properties of the file, you can compute stuff such as throughput, performance per MB etc.
Jan 27, 2012
Christoph Becker
minimee measures these attributes for some platforms (see http://www.ifs.tuwien.ac.at/~becker/pubs/lncs56480221.pdf
)
Of course, you want to relate the measures to each object to properly aggregate and analyse them over large experiments