Transparent and Flexible Network Management for Big Data Processing in the Cloud
summary
FlowComb is a network management framework that predicts Hadoop data transfers (detecting ~28% before they start and ~80% before they end), uses small software agents on every node to read logs and report sizes/status to a centralized decision engine, and enforces alternate routes by installing OpenFlow rules on a programmable network. On an 8-switch, 15-node testbed sorting 10GB it reduces completion time by ~35% while rerouting only ~6% of transfers; agent CPU overhead stays under 5%.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 0:35Motivation
- 2:00Partition Aggregation
- 4:32Flowcom
- 7:27Software Agents
- 8:11Transfer Scheduling
- 9:11Enforce Path Change
- 9:34Recap
- 10:19Preliminary Experiments
- 10:36Preliminary Results
related talks
16mNetwork Flow Data in the Cloud
23mNetwork Flow Aggregation: Pay for the Logs You Care About!
13mContracting Wide-area Network Topologies to Solve Flow Problems Quickly
26mSplit/Merge: System Support for Elastic Execution in Virtual Middleboxes
17mNetHint: White-Box Networking for Multi-Tenant Data Centers
26mFreeFlow: Software-based Virtual RDMA Networking for Containerized Clouds
15mForestColl: Throughput-Optimal Collective Communications on Heterogeneous Network Fabrics
29mFinding the Needle in a Haystack: Tenant-Level Network...
