Performance Optimization on Xen-based Android Device
summary
A Xen-based Android dom0 prototype runs Android almost natively by pinning vCPUs, passing through devices and PMU, and disabling unnecessary Xen scheduling, and most benchmarks reach about 97% of native. Using retune and xentrace plus perf/xen profiler identifies remaining hotspots: a storage IO gap falls from 21% to 8.5% after forcing low memory page cache and shrinking cache allocation, but a malware KPI stays at ~30–33% and home‑screen power shows a 1.1–1.2× overhead driven by pv-mmu overhead and TLS switching; HVM with EPT is proposed to close those gaps.
AI-written summary · about
chapters (11)
- 0:00Introduction and project goal
- 1:07Overview and design approach
- 5:02Design details and performance
- 7:07Interrupt and FPU overheads
- 9:01Power management strategies
- 10:47Suspend and wake-up mechanism
- 14:15Performance gaps identified
- 16:44Profiling and tuning tools
- 18:20Optimization case studies
- 25:36Future directions and summary
- 27:06Q&A session
related talks
48mXen and the Art of Embedded Systems Virtualization
38mBuilding Hard Real-Time Linux Systems with Xenomai
25mReducing Android Boot Times: Evaluating Proactive File Prefetching
33mNext Generation Xen Architecture for Automotive
35mThe Xenomai Project : Current and Future
53mXenpwn: Breaking Paravirtualized Devices
51mAdventures in (Simulated) Asymmetric Scheduling
41mSecuring Your Xen-Based Cloud