Volker Lendecke wrote:
On Sun, Mar 09, 2008 at 01:21:22PM -0700, Howard Chu wrote:
Howard Chu wrote:
I recall we had a few ideas last year, but they came up a bit late. Anyone still have some wish list items for this year?
I think it would be feasible for a student to develop a back-tdb backend. The basic backend functionality could be copied from back-hdb and the caching layers could be omitted, which would leave the code pretty straightforward. Since back-tdb is primarily an in-memory database it wouldn't be much of a performance issue, and it would get us something easy to configure for the very common, less demanding deployments.
Another possibility which I started discussing with Tridge is to modify the tdb code to always mmap to a fixed address space. In that case we could store Entry objects directly, with all pointers intact, which would eliminate the need for a deserialization step on Reads. Then the need for entry caching would be completely eliminated as well.
One thing that could also be kept in mind is a potential ctdb backend to go clustered. ctdb right now lacks transactions the way you would probably need them, but this is something that will be added.
Perhaps that should wait, and be a followon project then. Get a very basic tdb backend working first.