Distributed Transaction Server

Scalable fault-tolerant stock trading platform with Paxos consensus

Designed and implemented a scalable three-tier stock trading platform (Frontend, Backend — Catalog and Order services) capable of handling high user concurrency through a dynamic thread-pool scheduling model.


Some highlights of this project are:

  • Achieved redundancy and fault tolerance via Bully Leader Election and Paxos-based consensus, ensuring consistent state synchronization across replicas under node failures.
  • Enhanced frontend performance with an LRU caching layer that reduced latency for frequent stock lookup requests.
  • Stress-tested the system under high concurrency to validate throughput and consistency guarantees.

Advisor: Prof. Prashant Shenoy, UMass Amherst.


Figure: Client-server model by Calimo (derived from work by David Vignoni), LGPL, via Wikimedia Commons.