This course teaches participants about Fast One-sided RDMA-based Distributed Transactions for Disaggregated Persistent Memory. The learning outcomes include understanding how to improve resource utilization and failure isolation through persistent memory disaggregation, as well as how to design efficient distributed transaction schemes for disaggregated PM architecture. Participants will learn about leveraging one-sided RDMA, batching read and lock operations, updating remote replicas in a single round trip, and optimizing data persistency in the PM pool. The course is intended for individuals interested in distributed systems, memory disaggregation, and improving transaction throughput and latency in remote memory pools. The teaching method involves a theoretical overview, challenges, solutions, and evaluation of the proposed system.
FORD - Fast One-sided RDMA-based Distributed Transactions for Disaggregated Persistent Memory
Overview
Syllabus
Intro
Disaggregated Persistent Memory Memory disaggregation
RDMA-based Distributed Transaction
Transaction on Disaggregated PM
State-of-the-art
Challenge 1
Overview
Hitchhiked Locking Read and lock the read-write (RW) set in execution . Avoid subsequent locking and validations • No lock on the read-only data
Coalescent Commit Commit all replicas together in one round trip - In-place update: Parallel undo logging in execution phase • Prevent reading partial updates: Control data visibility
Selective Remote Flush Only issue remote flushes to backups after the final write • Ensure recoverability by backups • Reduce flushing round trips • Compatible with different flush primitives
Evaluation Benchmarks
Conclusion
Taught by
USENIX
