MIT Sloan AI Adoption: Build a Playbook That Drives Real Business ROI
Build the Finance Skills That Lead to Promotions — Not Just Certificates
Overview
Google, IBM & Meta Certificates — All 10,000+ Courses at 40% Off
One annual plan covers every course and certificate on Coursera. 40% off for a limited time.
Get Full Access
Learn about HDTX, a high-performance distributed transaction system designed for RDMA-based disaggregated memory architectures in this 19-minute conference presentation from USENIX ATC '25. Discover how memory disaggregation improves datacenter memory utilization and scalability while exploring the challenges of processing distributed transactions due to high remote memory access latency. Examine three novel design approaches: a fast commit protocol (FCP) that minimizes network round trips by coalescing transaction processing phases, an RDMA-enabled offloading mechanism that reduces data transfers between computing and memory nodes through strategic RDMA primitive orchestration, and decentralized priority-based locking for scheduling mission-critical transactions. Analyze experimental results demonstrating HDTX's performance improvements, including latency reductions of up to 88.3% compared to FaRM and 72.1% compared to FORD, along with throughput improvements of up to 2.08× and 84.7% respectively over existing RDMA-based distributed transaction systems.
Syllabus
USENIX ATC '25 - Fast Distributed Transactions for RDMA-based Disaggregated Memory
Taught by
USENIX