CS PhD student · University at Buffalo · distributed transactions & databases
I build and rigorously evaluate the systems behind strong consistency — distributed transactions, consistency protocols, and the datastores that run them. I like turning protocol theory into working systems: my code spans CockroachDB internals, TAPIR, Jepsen, and several self-built large-scale testbeds.
Shows that "replication-critical" read-only transaction designs are unnecessary: formalizes the SONIC and Remote-Read theorems and demonstrates them with two systems — Starry (a deep modification of UW's TAPIR, 300+ commits of C++) and an extended COPS-SNOW with chain replication — validated by self-built Jepsen falsifier suites. Latest SOSP submission advanced through two review rounds to PC-meeting discussion.
A transparent, proactive message-ordering layer inside gRPC interceptors that reorders transaction messages by timestamp before they reach data shards — cutting P99 latency by up to 75% and abort rates by up to 85% for strictly serializable concurrency control, with zero changes to database code.
An end-to-end framework that makes evaluating distributed storage systems easy, correct, and reproducible. I built its unified workload engine (YCSB, transactional YCSB-T, interactive TPC-C) and multi-language adapters for CockroachDB, Cassandra, and EPaxos-family protocols; our re-evaluation refuted EPaxos's published 3× throughput claim.
Directed implementation testing that turns community-known correctness pitfalls (e.g., timestamp inversion) into targeted tests — complementing random-exploration tools like Jepsen.
Always happy to talk distributed systems, databases, or LLM infrastructure. Reach me at sqiang@buffalo.edu or connect on LinkedIn.