Implement Distributed Shared Memory Over RDMA for a Trading Firm
Overview
What this challenge is about.
Build a DSM library in C++ exposing a region API (allocate, read, write, free). Use ibverbs for RDMA primitives (one-sided read/write, atomics for the directory). Implement directory-based sequential consistency: each page has a home node tracking the current owner. Run on a 4-node cluster (or QEMU + soft-RoCE). Benchmark with the firm's workload simulator (provided, anonymized): mixed read/write market-state access at roughly 200K ops/sec target. Deliver source, benchmark report covering latency distribution + invalidation overhead, and a 6-page memo recommending production prototype or not.
The Brief
What you'll do, and what you'll demonstrate.
Build a DSM-over-RDMA prototype with sequential consistency and assess production viability against a real trading-workload simulator.
Earning criteria — what you'll demonstrate
- Understand RDMA verbs and one-sided operation semantics
- Implement a directory-based coherence protocol from scratch
- Reason about sequential consistency vs weaker memory models
- Benchmark distributed memory systems with realistic workloads
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Career mappings coming soon.