C. Ciprian (Romania)
Distributed Databases, Algorithms
The increased size of databases does not allow us any more to put everything on the same computer, so, for these databases, we are forced into a distributed environment. The number of replicas of the data is directly proportional to the time it takes to do a write, so, by increasing the number of the databases, we eventually (and quite early) end up increasing the write time beyond an acceptable threshold. Since the user can not wait until we update all of our replicas, we are forced to return control to the user before all the writes are completed, therefore leaving the database in an inconsistent state. Another big reason for not requiring all the replicas to be updated in a single transaction is availability - we want our application to remain responsive in partial failure scenarios. We present a new variation of session guarantees(Terry et al. [6], Brzezinski et al. [2]) and describe our experiences in building a middleware layer for these types of guarantees: JStabilizer([3]). Sim ilar approaches were taken by Mastroianni et al. in [5] in the field of Individual Knowledge Workers.
Important Links:
Go Back
