PLOVER: Fast, Multi-core Scalable Virtual Machine Fault-tolerance
summary
PLOVER implements Virtualized SMR to replicate VMs with three replicas (primary, backup, witness), using Paxos for total order and RDMA to write requests and compute per-page hashes so only divergent pages are transferred. Evaluation shows 72–97% identical dirty pages so PLOVER sends far fewer pages, achieving 2.2X–3.8X throughput and 9.2X less network bandwidth than popular primary-backup systems. It detects low-activity sync points via hypervisor CPU/disk stats and publishes code at github.com/hku-systems/plover. RDMA page-hash comparison
AI-written summary · about
chapters (8)
- 0:00Introduction to VM faults
- 1:03Limitations of Remus
- 4:40State machine replication
- 6:30Plover design overview
- 8:24Synchronization mechanisms
- 10:23Optimization techniques
- 12:54Evaluation and results
- 19:10Conclusion and Q&A
related talks
23mTeRM: Extending RDMA-Attached Memory with SSD
45mBuilding Available and Partition Tolerant...
12mCharacterizing and Optimizing Remote Persistent Memory with RDMA and NVM
27mStatus Update of COLO Project
39mTrust Deterministic Execution to Scale & Simplify Your Systems
17mPragh: Locality-preserving Graph Traversal with Split Live Migration
16mFORD: Fast One-sided RDMA-based Distributed Transactions for Disaggregated Persistent...
01h 09mAlgorithms for practical distributed agreement