Adaptive Transaction Processing with Reinforcement Learning

Location

New Computer Science-1-Room 120 (105 Seats) (105)

Event Description

Please join us for the next CSE 600 Seminar this Friday, October 11th, at 2:30pm in New Computer Science 120 given by Assistant Professor Mohammad Javad Amiri.

Abstract:

Today's distributed transaction processing systems must deal with untrustworthy environments where multiple mutually distrustful entities communicate with each other, and maintain data on untrusted infrastructure. Byzantine Fault-Tolerant (BFT) protocols have recently been extensively used by distributed transaction processing systems to establish consensus on the order of transactions. However, the proliferation of different BFT protocols has made it difficult to navigate the BFT landscape, let alone determine the protocol that best meets application needs. Moreover, as novel smart contracts, modern hardware, and new cloud platforms arise, future-proof distributed transaction processing systems need to be designed with full-stack adaptivity in mind. This talk presents our vision for a reinforcement learning (RL)-based distributed transaction processing system that adjusts effectively in real-time to changing fault scenarios and workloads.

Date Start

Date End