Performance Evaluation of Live Migration based on Xen ARM PVH
summary
Implemented live migration for Xen ARM PVH on ARM boards and measures real effects with a streaming workload: dual‑core virtualized board supports ~110 concurrent clients (about 7 clients per watt), a quad‑core board reaches ~300 clients at 18.9 W (~15 CPW), and consolidation shows ~70% energy reduction with 4‑to‑1 packing. Dirty‑page tracking uses page‑table walking vs a virtual linear page table (much faster), temporary linked‑list logging, ~1 s downtime, but total migration can hit ~880 s when network saturates.
AI-written summary · about
chapters (18)
- 0:00Intro
- 0:30Contents • Motivation
- 1:10Energy Problem in Datacenters • Datacenters eat up magnificent amount of electricity
- 1:50ARM Servers for Future Green Data Center
- 2:23Overall Architecture • Components for Live Migration in Xen ARM PVH
- 3:13Sequence of Live Migration
- 5:36Dirty-page Tracing: Dirty-page Detection
- 7:45Two choices for finding PTE
- 7:57Manual Walking of p2m Table
- 10:02Dirty-page Tracing: Get-dirty Bitmap
- 12:46Scenarios
- 14:11Maximum capacity of ARM Arndale
- 14:57Energy-efficiency comparison to x86 Server
- 15:48Case 1: Quad-core ARM board (In-progress) • ARM board: 4 ARM cores with 8GB memory
- 17:13Case 2: Streaming Server Consolidation of Xen ARM Arndale
- 18:43Case 2: Total time of live migration . Migrate a VM at a time - With different domu memory size 128MB
- 19:50Case 2: Live Migration Performance • Number of dirty-pages in iterations
- 21:44Questions?
related talks
48mXen and the Art of Embedded Systems Virtualization
33mNext Generation Xen Architecture for Automotive
38mBuilding Hard Real-Time Linux Systems with Xenomai
42mEnhancing Hypervisor and Cloud Solutions Using Embedded Linux
31mDebugging Tools and Techniques for Virtualized/Containerized Automotive Systems
18mZVM: Zephyr-based Virtual Machine
36mShared Virtual Addressing for High Performance Arm Infrastructure Platforms
40mZero-Overhead Virtualisation: It's a Trap!