MARS: Replicating Petabytes over Long Distances
summary
MARS Light is a kernel‑level asynchronous block‑replication module and accompanying user tools that replicates petabytes across long‑distance links. It buffers gigabytes of changes, tolerates packet loss and high latency, and provides transactional write‑log semantics with time‑consistency guarantees. The system is deployed in production (2×6 PB across hundreds of servers) and lab and live results demonstrate its operation.
AI-written summary · about
chapters (11)
- 0:00MARS introduction
- 3:21Long-distance replication
- 8:13DRBD vs MARS
- 14:40How MARS works
- 23:14Disaster recovery with MARS
- 30:59Metadata consistency
- 36:10MARS use cases & scaling
- 41:06Q&A
- 54:08MARS docs & future
- 59:15Performance optimizations
- 1:05:02Summary
related talks
23mEasy Geo-Redundant Handover + Failover with MARS + systemd
43mHot Deploying Low-Latency Services for 24/7 Operation
01h 19mBuilding an Efficient Replication System for Thousands of Terminals
43mMaintaining Consistency in a Massively Parallel Environment
42mUsing Mesos to Drive Devops Adoption at Scale
18mLeveraging Multiple Regions to Improve Site Reliability:
39mHow we learned to love the Data Center Operating System
31mHow Millennium Bcp Leverages Radius To Empower Developer + Operator Collaboration