System::getStatistics() and Objects created
Posted: Fri Aug 07, 2009 12:20 pm
by JensRichnow >> Tue, 4 Feb 2003 20:22:23 GMT
I want to capture the amount of objects created for a certain process. I tried varies approaches, i.e., use the profiler (process.startProfile(), etc.), use the System::getStatistics() or Monitor and attempt to write my own sampler.
You might guess all three outputs vary considerably, although I would have thought that the two methods provided in Jade would be at least close.
Does somebody know whether the "Objects Created" parameter in the first two sampling methods includes persistent as well as (shared) transients?
In my own sampling code I can easily differentiate the creation mode (persistent, vs. (shared) transient). Currently I put the code in the ::create() of as many superclasses in the schema as possible and that gives me a pretty good picture. It would be nice if one could somehow attach the sampling to the Object class as this would be the sensible place to do so. The sampling currently does not include collections created exclusively and is therefore 'undersampled'.
I know that the Jade provided methods include all objects created (system-wide statistics).
I'm concerned for two reasons. (1) The Monitor for instance gives me at least twice the amount of objects created as the 'hand sampling'. (2) In this particular application I have to multiple the objects created for one process by at least 100,000. Therefore a few objects in the sampled process will have quickly repercussions on the DB performs I would assume.
I want to capture the amount of objects created for a certain process. I tried varies approaches, i.e., use the profiler (process.startProfile(), etc.), use the System::getStatistics() or Monitor and attempt to write my own sampler.
You might guess all three outputs vary considerably, although I would have thought that the two methods provided in Jade would be at least close.
Does somebody know whether the "Objects Created" parameter in the first two sampling methods includes persistent as well as (shared) transients?
In my own sampling code I can easily differentiate the creation mode (persistent, vs. (shared) transient). Currently I put the code in the ::create() of as many superclasses in the schema as possible and that gives me a pretty good picture. It would be nice if one could somehow attach the sampling to the Object class as this would be the sensible place to do so. The sampling currently does not include collections created exclusively and is therefore 'undersampled'.
I know that the Jade provided methods include all objects created (system-wide statistics).
I'm concerned for two reasons. (1) The Monitor for instance gives me at least twice the amount of objects created as the 'hand sampling'. (2) In this particular application I have to multiple the objects created for one process by at least 100,000. Therefore a few objects in the sampled process will have quickly repercussions on the DB performs I would assume.