Persistent Memory plus RDMA, new age remote device
summary
Persistent memory offers DRAM-like latency plus nonvolatility, but plain RDMA lacks guarantees for durability and atomic updates; this work implements RDMA flush and RDMA atomic write semantics in software using soft RoCE plus extended ibverbs and ships an up‑layer library librpma that emulates flush via RDMA read or send/recv (handles Intel DDIO) and atomic write via ordered writes. Implementation runs in soft RoCE, hooks into kernel rdma core and a fio engine for benchmarks; remote PMEM throughput/latency is slightly worse than local PMEM but functional and measurable.
AI-written summary · about
chapters (7)
- 0:00Introduction and agenda
- 1:26Challenges of RDMA
- 6:33RDMA specifications
- 9:27Soft RoCE implementation
- 18:41Remote memory library
- 25:31Performance evaluation
- 27:04Conclusion and future work
related talks
18mFileMR: Rethinking RDMA Networking for Scalable Persistent Memory
15mReplicating Persistent Memory Key-Value Stores with Efficient RDMA Abstraction
25mDisaggregating Persistent Memory and Controlling Them Remotely
14mODRP: On-Demand Remote Paging with Programmable RDMA
18mRDMA is Turing complete, we just did not know it yet!
12mPreview: Networking
20mPeRF: Preemption-enabled RDMA Framework
16mAchieving Wire-Latency Storage Systems by Exploiting Hardware ACKs