Leveraging SQL for complex transactions on modern hardware platforms


  Guerraoui Rachid

Research Partners

Swiss Joint Research Center - Microsoft SJRC - Microsoft

In 2017 research leads involved in the Swiss Joint Research Center (Swiss JRC) attended a workshop at the Microsoft Research Cambridge Lab to mark the launch of a new phase in the collaboration with Microsoft. Apart from Microsoft, the two primary stakeholders in the Swiss JRC are ETH Zurich and EPFL. On the agenda was the project titled “Revisiting Transactional Computing on Modern Hardware,” one of six EPFL projects approved by Microsoft.

Project Website

The principal investigators of this project are Rachid Guerraoui and Georgios Chatzopoulos of EPFL, and Aleksandar Dragojevic of Microsoft Research Cambridge lab.

The project is extremely relevant in today’s computing world, where memory needs have far outpaced the capacity of machines; it is no longer feasible to store all the memory on a single machine, necessitating distribution and synchronization of data across a group of machines. However, scaling up transaction systems is a resource-sucking process because augmenting the number of processors and memory, and adding a robust server, could cost a hefty amount. Besides, there is always the risk of failure and outages. Conversely, scaling out distributed transaction is a much cheaper option for production systems.

Considering this aspect, the researchers plan to leverage Structured Query Language (SQL) to design and provide support for complex transactions on top of modern hardware platforms. The premise of the research is that the straightforward implementation of SQL support will not contribute to improved performance of transactions. Therefore, the focus of the research will be on revisiting the design of abstractions to determine the best method to maximize the benefits of modern networking and hardware capabilities.

The researchers have chalked out the schema for research, and expect the outcome to have a major impact on transactional computing with modern platforms such as Fast Remote Memory (FaRM).

Suggested reading: