1 / 16

RT- Xen : Towards Real-Time Hypervisor Scheduling in Xen

RT- Xen : Towards Real-Time Hypervisor Scheduling in Xen. Sisu Xi, Justin Wilson, Chenyang Lu, and Christopher Gill. Background – Virtualization . System Integration, Reducing Cost, etc. App. App. App. App. R eal-time aware. OS. OS. Xen Hypervisor. Not real-time aware. Hardware.

curry
Télécharger la présentation

RT- Xen : Towards Real-Time Hypervisor Scheduling in Xen

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. RT-Xen: Towards Real-Time Hypervisor Scheduling in Xen Sisu Xi, Justin Wilson, Chenyang Lu, and Christopher Gill

  2. Background – Virtualization • System Integration, Reducing Cost, etc. App App App App • Real-time aware OS OS Xen Hypervisor • Not real-time aware Hardware Hardware Hardware Round-Robin fashion NO priority between OS Time Problem: Real-Time Applications can NOT benefit from Virtualization

  3. Motivation – From Scheduling’s View App App App App App App App App … Leaf Sched OS Sched OS Sched Leaf Sched … Xen Scheduler Root Scheduler Leaves are implemented as Servers (Period, Budget, Priority) Solution: Incorporate Hierarchical Scheduling

  4. Contributions – RT-Xen • First real-time hypervisor scheduling framework for Xen • Most widely used open-source virtual machine monitor (VMM) • Open-source platform for researchers to develop and evaluate real-time hierarchical scheduling techniques • Instantiates four fixed-priority servers • Deferrable, Periodic, Polling, Sporadic Server • First comprehensive experimental comparison of these servers within a same virtualization platform

  5. Server Design – Deferrable & Periodic • Servers (Period, Budget, Priority) T1 (10, 3) S1 (5, 3, 1) with Two Tasks T2 (10, 3) Time 2 15 0 10 5 Actual Execution back-to-back Deferrable Server 3 Budget in S1 Time 0 10 5 15 2 Actual Execution IDLE theory favored Periodic Server 3 Budget in S1 Time 2 0 10 5 15

  6. Implementations • Implemented in Xen 4.0 • All source code available at http://sites.google.com/site/realtimexen/ • Support only one VCPU per domain (single core) • VCPU (period, budget, priority) Xen Scheduling Framework Credit Scheduler SEDF Scheduler RT-Xen Sub-Framework Deferrable Server Periodic Server Polling Server Sporadic Server

  7. Experimental Setup Scheduling Algorithm (Deferrable, Polling, Periodic, Sporadic) (Period, Budget, Priority) for Dom1 (Period, Budget, Priority) for Dom2 … Use Rate Monotonic within each Domain For each task: shorter period -> higher priority IDLE App App App App Dom0 Dom1 Dom5 … VCPU VCPU VCPU RT-Xen Schedulers (Deferrable, Polling, Periodic, Sporadic) HW Core 0 Core 1

  8. Overhead context switch Five tasks per Domain, Five Domains System Load: 70% do_schedule() Time Run for 10 seconds +0.02% 1. Context Switch >> do_schedule() 2. Acceptable overhead to many real-time systems 3. Differences between server algorithms are insignificant

  9. Soft Real-Time Performance – Setup High • VCPU Share = Budget / Period • System Load: 30% -> 100% +5% Priority • Credit: Same Share Decreasing Even Increasing easiest common hardest

  10. Soft Real-Time Test Decreasing Increasing Deadline Miss Ratio Total System Load

  11. Soft Real-Time Test - Credit Decreasing Increasing Deadline Miss Ratio Theory Guarantee When Deadline Miss Credit scheduler can NOT deliver real-time performance. Total System Load

  12. Soft Real-Time Test - Deferrable Decreasing Increasing Deadline Miss Ratio [30%, 45%] [R. Davis, A. Burns, RTSS 2005] Total System Load

  13. Soft Real-Time Test - Periodic Decreasing Increasing Deadline Miss Ratio [30%, 50%], [60%, 75%] [R. Davis, A. Burns, RTSS 2005] Theory is pessimistic. Total System Load

  14. Soft Real-Time Test - Overloaded Decreasing Increasing Deadline Miss Ratio Deferrable Server performs well Periodic Server performs worst Total System Load

  15. Summary • Instantiation and empirical study of fixed-priority servers • Deferrable Server, Polling Server, Periodic Server, Sporadic Server • First comprehensive experimental comparison of these algorithms in the same virtualization platform • Overhead differences between servers are insignificant • For soft real-time performance: • Theory is pessimistic • Deferrable Server generally performs well • Periodic Server performs worst under overloadedsituations

  16. Pointers • RT-Xen Website: • https://sites.google.com/site/realtimexen/ • All source code including test programs, task sets, documentation. • Call for users and contributors!

More Related