From: Elias R. <er...@m-...> - 2006-04-02 20:50:40
|
On Fri, 2006-03-31 at 18:46 -0500, Thompson, Bryan B. wrote: > I think that I need some more background on what your application is trying > to do before I can offer any specific advice. Caching for jdbm resolves > recently access persistent records against those in the cache. Are you > getting the same behavior from another layer? Yes, in fact the point of JBossCache is to provide a memory layer. :-) By the way, I'm getting much better results through a batching process I wrote. Thanks for helping me out! > The reason for the various serializer interfaces is that historically > serialization was for simple records only and there continues to be a use > case for that. The stream based stuff is more recent and was driven by the > need to support frameworks layered over jdbm. I know there are concerns on backwards compatibility, though if you're considering adding new interfaces, I think one solid interface would be sufficient. By the way, for browsing BTree nodes... It would be nice if there was a browse option just for keys which didn't load the values. I'm thinking you could create a custom deserializer that did not read in the BTree values, just the keys. To do this, just from browsing the code, it seems like you'd have to make some internal changes to return InputStream rather than byte[], e.g. add a PhysicalRowIdManager "fetchStream" or something. Would this be something I could contribute? This would certainly be a memory improvement. |