Hi all,
I’m ruminating on a wholesale performance assessment. Looking around I think we have:
Fixlets for the FillDB performance log:
- BES Support: 4778 Enable FillDB performance log
- BES Support: 4779 WARNING: FillDB performance log is enabled
Via the MXPerformanceToolkit, we have MXPerfmon[.exe], which is a shim for logman.exe (on Windows; aka the cli interface for Performance Monitor aka perfmon.msc) and nmon` (on Linux).
On Windows, it seems that logman.exe has options for specifying start/end times for a named counter, and can use XML configuration files. While MXPerfmon.exe has options to display the configuration items, it doesn’t close the circle for start/end times, nor supply the bits for XML output.
Meanwhile, there’s an older KB article on manually creating Windows logman counters.
If one wants to have a coordinated performance sampling across the infrastructure (on Windows), it looks like we need to:
- Deploy the FillDB performance log fixlets, per BF scheduling.
- Configure
logmancounters with schedules.- For consistency
- Use
MXPerfmonto output the list of configuration items. - Feed that text to
logman.exe, with additional options for start/end, exact names, etc, and output the XML file(s). - Distribute the XML files to infrastructure servers
- run
logman importwith the XML file(s) - [This sequence could be orchestrated via custom fixlets]
- Collect the data
- UploadManager?
- Run around and manually copy files?
And then, finally, analyze the output.
Am I missing anything? Anyone have better strategies?
Thanks!
