Forums

by JensRichnow >> Mon, 9 Dec 2002 20:32:03 GMT

As we are about to implement a subsystem which statistically analyses the usage pattern of the various applications against a multitude of parameters etc. I wondered what a good approach would be.

Currently I evaluate two options: (1) Have a background process record all required data via notifications (persistent transactions), and (2) write the data to structured files, and process the files at request time, perhaps on a weekly basis. As all current applications are kind of optimised for speed we want to keep the overhead of capturing the additional data to a minimum.

What other options could be used? How does the persistent transaction overhead (time-wise) compares to writing to the file system? Does anybody did some rough benchmarking which compare the above two options?

by allistar >> Mon, 9 Dec 2002 21:06:35 GMT

We do the same thing (each operation a user does - load, save, delete)
we capture the statistics of the database node (object gets, locked, unlocked, total time etc.). I store these statistics in a persistent object, and to minimise the impact on the system I don't place them in
a collection. When they get output to a file (every month or so) I iterate through instances of the class.

I'm not sure if these are the sorts of stastics you are talking about, but my gut feel is that storing them in persistent objects in the database would be the way to go. As long as you are not updating a
single collection (which could give locking problems) there should be negligible impact on performance.

Allistar.

------------------------------------------------------------------
Allistar Melville
Software Developer
Auckland, NEW ZEALAND

Greentree International,
Developer of Greentree Financial Software. ------------------------------------------------------------------