Tintin: A Unified Hardware Performance Profiling Infrastructure to Uncover and Manage...
summary
Tintin is a Linux-kernel PMU subsystem with three modules—monitor, scheduler, manager—that quantifies multiplexing error at runtime, schedules events to minimize total uncertainty, and exposes uncertainty via an OS primitive Event Profiling Context (ePX). On benchmarks Tintin keeps error below 5% when tracking >100 events versus >20% for native perf, adds ~0.5 percentage points overhead (perf 1.9% → Tintin 2.4% average), beats round‑robin via elastic scheduling in 64% of cases, and improves real workloads (outperforms IMA in 95% of Pond runs; adding uncertainty helps models in 55% of cases).
AI-written summary · about
chapters (8)
- 0:00Tintin introduction
- 1:43Hardware counter limits
- 4:57Design insights
- 7:31Tintin's components
- 13:16Reporting uncertainty
- 15:04Evaluating accuracy
- 16:54Pound case study
- 19:02Summary & Q&A
related talks
45mLinux perf tool metrics
59mBeyond pprof, Hardware Performance Counters
45mPerformance lab. The power of the perf tool!
41mLinux Systems Performance
51mLinux Systems Performance - Architecture
28mLevel Up Your Game: OS Kernel and Game Interactions revealed with Perfetto
16mXkernel: Principled Performance Tunability of Operating System Kernels
35mIntegrated PCIe monitoring and tracing facilities