Comments?
The last time I tried this many years ago the thing floundered on the details of isolation. So you might watch out for that : clients will want to see varying levels of isolation depending on their goals. For example a client may want to see the effects of their updates reflected in searches that they perform within the same transaction. OTOH if one implements serializable isolation semantics then I believe it is impossible to avoid deadlocks and hence clients need to be taught how to detect them, back off and re-try. Follow this line of thought for a while and one (well, at least myself and the folk that were working on the project at the time) concludes that anyone needing transactions should use a RDBMS and LDAP is better off without them ;) My draft is probably somewhere out there in the googlesphere somewhere - I seem to remember it being similar to yours... Ah, here it is : http://www.openldap.org/lists/ietf-ldapext/199804/msg00046.html The subsequent thread brings up some other points that I'd since forgotten.
