Scalable Rule Management for Data Centers
summary
A system called vCRIB automates rule placement across hypervisors and switches using a two-step pipeline: a resource-aware feasible-placement (a firstfit-decreasing-similarity algorithm that packs similar partitions) followed by a traffic-aware refinement (a benefit-greedy migration). Evaluated on a 1,000-server simulation with 200k flows, it finds feasible placements when source-only placement would exceed host capacity and often keeps extra forwarding under <10% traffic overhead, extending operational range of data centers.
AI-written summary · about
chapters (10)
- 0:00Intro
- 0:16Introduction: Definitions
- 1:38Current practice
- 1:51Machines have limited resources
- 4:05Challenges: Concrete example
- 6:02Contribution: VCRIB, a Virtual Cloud Rule Information Base Proactive rule placement abstraction layer
- 6:59Partitioning with replication
- 10:39Traffic-aware refinement Overhead greedy approach
- 12:29VCRIB design: Dynamics
- 16:05Parameter sensitivity analysis: Rules in partitions
related talks
16mFlow Scheduling with Imprecise Knowledge
14mRDC: Energy-Efficient Data Center Network Congestion Relief with Topological
25mResQ: Enabling SLOs in Network Function Virtualization
15mMitigating Scalability Walls of RDMA-based Container Networks
25mVirtualizing Traffic Shapers for Practical Resource Allocation
17mPrecise Data Center Traffic Engineering with Constrained Hardware Resources
20mContra: A Programmable System for Performance aware Routing
23mMinimal Rewiring: Efficient Live Expansion for Clos Data Center Networks
