Manuel Bravo, Post-doctoral Researcher at the IMDEA Software Institute has proposed elegant algorithms for transaction processing in FARM-like databases, proved their correctness and identified how they should be changed to exploit Remote Direct Memory Access (RDMA), a new networking technology that has the potential to drastically increase the scalability of distributed databases. However, to realize this potential, the algorithms for transaction processing in such databases have to be radically redesigned.
The first industrial database to use RDMA is the recent FARM system from Microsoft, which is now used to compute some of the search queries in Microsoft Bing. Unfortunately, the theoretical foundations of FARM have been shaky: its algorithms have not been proven correct and the rationale many of its design decisions was unclear.
Bravo, that has presented this proposal as a paper to PODC, demonstrates that the methodology for modifying transaction processing algorithms to run over RDMA also opens the door to exploiting RDMA for scaling up other industrial databases. This work has increased confidence in the correctness of FARM design, and allowed proposing its simplifications,
