Large-Scale Distributed Transactions
Transactional Memory (TM) is a synchronization mechanism dedicated to the multicore era where cores communicate through shared memory. Recent results have shown how TM can leverage cache coherence protocols, this led to the definition of a new concept called Distributed Transactional Memory (DTM). Although priority-based contention managers have been extensively studied in the shared memory model, there is very little work on their applicability in message-passing model.
In a large-scale network where nodes can issue transactions on distributed objects, the cost of local computation is negligible compared to the communication cost. Hence, contention management in DTM has to take communication cost into account to map priority to transactions.
The goal of this project is:
- to apply traditional contention managers to the context of DTMs in large-scale networks,
- to implement some contention managers using a Peer-to-Peer simulator,
- to conclude based on the obtained simulation results.
Responsible: Vincent Gramoli
Project intended for 1 student