Comment on ContainerCache?

[prev] [thread] [next] [Date index for 2004/07/16]

After reading about the LiveObjects in CDBI and idea sprang into my mind 
(well, more an adaption from EJB container) and before I start 
implementing it, I would here comments about it. So idea is as follows:

Similar to LiveObjects, I would apply a ContainerCache per class/table. 
When retrieving a object/row, it would first look in the CacheContainer 
and only when it doesn't exists there, it would connect to the database 
and put it into the ContainerCache.

The difference to LiveObjects is the lifetime of objects in the 
ContainerCache, which would be much longer than in LiveObjects, i.e.
a) as long as the low-watermark hasn't been reached, objects will live 
forever
b) above the high-watermark, all objects with refcount < 1 (not 
currently referenced in the rest of the program) will be removed until 
the low-watermark is reached


The main disadvantage I currently see:
The cycling through the ContainerCache and testing for refcount < 1 
might take some time.
The program will need more memory (depending on the high-watermark)

The main advantage:
* Much less round-trips to the database
* 'LiveObject' like features without dependence on 
garbage-collector/weak-references (identical to LiveObject when 
high-watermark = 0)
* General and centralized caching system (Currently, I'm usually caching 
small (<100 rows) table completely, and for large tables I make sure 
that I keep my rows long enough alive so that I skip roundtrips.)


Comments are very welcome, in particular since it will be very hard to 
implement it as Plugin.

Best regards,

Heiko