Tuning Memory and the Network Stack in Linux
summary
Migration of high-throughput video delivery servers (each pushing ~40 Gbit/s) from OpenSuSE 10.2 to CentOS 7 exposed issues: unexplained swapping and OOM killer runs, uneven CPU core load, connection drops, and CPU load spikes. The talk explains how those problems were diagnosed and mitigated by tuning Linux memory/subsystem behavior and the network stack, describing the specific tools and kernel/settings changes used.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:51Video server issues
- 2:15Memory fragmentation fix
- 6:57Swap and NUMA problem
- 11:17Network tuning and broken pipe
- 14:34Network load balancing
- 18:48RSS, RPS, XPS, RFS
- 20:57Accelerated RFS and Intel
- 22:45Broken pipe consequences
- 23:07Softirq CPU load
- 27:42Gradual changes
- 29:11Summary
- 29:37Q&A
related talks
55mPerformance tuning of Red Hat Enterprise Linux
51mLinux Systems Performance - Architecture
52mPerformance tuninng for Red Hat Enterprise Linux.
01h 36mTutorial: Linux Memory Management and Containers
23mBuilding a Network Operating System Using Linux and Yocto
27mDebugging Linux Issues with eBPF
01h 36mLinux Memory Management and Containers
25mShenango: Achieving High CPU Efficiency for Latency-sensitive Datacenter Workloads
