jdbm

Kevin,

I was wondering if you had some means to cluster access in your application. Without that I do not think that this approach is going to win. There are two places in which it could be expected to outperform the recman: (1) clustered access to application objects; and (2) the recman cache for a given size of the cache. However it simply has to require more record retrievals to locate something using the BTree that it does to retrieve a given record using the record manager - simply because the BTree is using the record manager for its own pages, so you need to walk the btree which means multiple trips through the record manager.

Also, I expect that you could do this with Unicode keys, e.g., URIs, if you have support for compressed keys in the primary key index (this is not something that I have tried to get working with the current cvs source).

-bryan