NrOS: Effective Replication and Sharing in an Operating System
Talkat 15th USENIX Symposium on Operating Systems Design and Implementation (OSDI '21) · 2021-07-14..07-16·
summary
NrOS implements a sequential kernel per NUMA node and keeps a replica per node, synchronizing replicas with an operation log with combiner batching so cores read locally and one combiner applies batched writes. Results show scaling to 96 cores, memcached +31% throughput and 36% fewer cycles on page walks, and up to ~100× better than Linux on a frame-insert benchmark. A CNR extension uses multiple logs by commutativity (partitioning by PML4 entries) to scale concurrent updates; costs include extra replication work, memory overhead, and tricky TLB/shootdown handling.
AI-written summary · about
chapters (5)
related talks
16mOneSidedMW: Managing Disaggregated Memory Efficiently, Flexibly, and Securely with RNIC
15mLODIC: Logical Distributed Counting for Scalable File Access
17mSRNIC: A Scalable Architecture for RDMA NICs
20mExploiting Commutativity For Practical Fast Replication
16mState-Compute Replication: Parallelizing High-Speed Stateful Packet Processing
01h 09mAlgorithms for practical distributed agreement
24mPLOVER: Fast, Multi-core Scalable Virtual Machine Fault-tolerance
27mStatus Update of COLO Project