Analysis of interrupt latencies in a real-time kernel
2 likes3,402 views
This document analyzes interrupt latencies in a real-time kernel. It discusses how real-time operating systems must ensure operations complete within fixed deadlines to maintain predictability. When dealing with device drivers, this implies managing race conditions and fulfilling temporal constraints. The document evaluates the performance of the Linux real-time preempt patch using cyclictest to measure latency and compares its performance to the standard Linux kernel.
Analysis of interrupt latencies in a real-time kernel
- 1. Analysis of interrupt latencies in a real-time Università degli Studi di Trento kernel Facoltà di Scienze Matematiche Fisiche e Naturali Real-time operating systems Real-time Linux CFS Analysis of interrupt latencies in a Recap real-time kernel March 26, 2008 Laureando: Gabriele Modena (Mat. 108742) Relatore: dott. Luigi Palopoli Correlatore: dott. Luca Abeni
- 2. Analysis of Real-Time Operating Systems (RTOS) interrupt latencies in a real-time kernel Real-time operating systems Real-time Linux CFS Recap Ensure that an operation completes within a fixed amount of time (deadline) =⇒ predictability. When dealing with device drivers this implies managing race conditions; fulfill temporal constraints. This thesis focuses on shared mixed resources.
- 3. Analysis of Latency interrupt latencies in a real-time kernel Real-time Latency is the measure of the difference between the operating systems theoretical (algorithm) and actual (implementation) Real-time Linux schedule. CFS Recap In a RTOS latency: should be bounded; the bound should not be too high. The goal of a real-time kernel is to minimize latency by using algorithms and technology that properly implements them: preemptible kernel.
- 4. Analysis of Linux as an RTOS interrupt latencies in a real-time kernel Real-time operating systems Linux vanilla: Real-time Linux since 2.6 preemptible kernel; CFS The rt-preempt patch: Recap developed by Ingo Molnar; makes Linux hard real-time; uses the new (2.6.21) high resolution timers; preemptible critical sections; real-time mutexes; threaded interrput handlers.
- 5. Analysis of Benchmarking rt-preempt interrupt latencies in a real-time kernel Real-time operating systems Real-time Linux CFS Recap Two goals: test the actual real-time performance; test the performance of I/O scheduling. We look for a trade-off between latency and throughput.
- 6. Analysis of Cyclictest interrupt latencies in a real-time kernel Real-time operating systems Real-time Linux CFS high resolution test program written by Thomas Recap Gleixner; measure latency; schedules periodic threads; When it comes to real-time applications rt-preempt gives the expected performance.
- 7. Analysis of Vanilla vs. rt-preempt (1) interrupt latencies in a real-time kernel Probability of the latency (num. of samples / total samples) Real-time Preemptible Linux operating systems 1 Real-time Linux CFS 0.1 Recap 0.01 0.001 0.0001 1e-05 1e-06 0 1000 2000 3000 4000 5000 6000 Latency (usecs)
- 8. Analysis of Vanilla vs. rt-preempt (2) interrupt latencies in a real-time kernel Probability of the latency (num. of samples / total samples) Real-time RT Linux operating systems 1 Real-time Linux CFS 0.1 Recap 0.01 0.001 0.0001 1e-05 1e-06 0 10 20 30 40 50 60 Latency (usecs)
- 9. Analysis of I/O scheduling interrupt latencies in a real-time kernel Real-time operating systems Real-time Linux CFS In traditional operating systems interrupt handler Recap have priority above everything. rt-preempt =⇒ soft and hard irq handlers are threads: scheduling I/O ≡ scheduling tasks.
- 10. Analysis of Benchmark network throughput interrupt latencies in a real-time kernel Real-time operating systems Real-time Linux CFS netperf; Recap UDP traffic; 750B datagrams; schedule the real-time network threads; 30 minutes execution.
- 11. Analysis of Fixed priority scheduling interrupt latencies in a real-time kernel It is not possibile to obtain low latency and high throughput; Real-time operating systems Real-time Linux Net priority min. lat (µs) max. lat (µs) Tp (Mbps) CFS 1 20 94 38.9 Recap 50 22 96 39.9 80 20 148 77.4 99 22 164 73.4 Table: Linux 2.6.24-rc2-rt1, 750B UDP datagrams high throughput and high latency; low throughput and low latency. Fixed priority scheduling is not suitable =⇒ we look for a method to assign CPU quota to tasks.
- 12. Analysis of Completely Fair Scheduler interrupt latencies in a real-time kernel Real-time operating systems Real-time Linux CFS Recap implemented by Ingo Molnar; merged in mainline since 2.6.23; weights tasks by ratio; group scheduling;
- 13. Analysis of cfs & rt-preempt interrupt latencies in a real-time kernel My work Real-time use CFS and modify the nice system call to set operating systems weights (> 200 µs); Real-time Linux CFS use CFS and tune Control Groups (> 1000 µs). Recap Results net cyc load lat (µs) Tp (Mbps) 1 1 1 2931 49.9 2 1 1 3911 67.3 2 2 1 3721 71.1 4 2 1 6117 80.9 1 1 4 2871 35.2 Table: Scheduling real-time threads with CFS
- 14. Analysis of cfs & rt-preempt interrupt latencies in a real-time kernel Real-time operating systems Real-time Linux CFS impact on the I/O throughput =⇒ the ratio system Recap works; high latencies. Short time possible solutions CFS clean up and tuning (in progress); EDF scheduling + time limit (Linux 2.6.25).
- 15. Analysis of Overview of the work interrupt latencies in a real-time Step 1 kernel litterature study on real-time computing; Real-time study of the rt-preempt patch and real-time under Linux; operating systems Real-time Linux Step 2 CFS Recap benchmark real-time performance; study the relationship between I/O and real-time applications; Step 3 test the CFS scheduler; try custom and standard solutions; Future work results presented at the RTAS 2008 conference (http://www.rtas.org/ ); tuning CFS and testing new scheduling policies.
- 16. Analysis of Questions interrupt latencies in a real-time kernel Real-time operating systems Real-time Linux CFS Recap Any question?