The document discusses RT-Xen, a real-time virtualization framework designed for efficient scheduling in Xen hypervisors, emphasizing applications in automotive systems and cloud gaming. It details various scheduling policies, including global and partitioned scheduling, and outlines the performance guarantees provided to virtual machines (VMs) with real-time demands. The framework allows for the implementation of compositional scheduling theory, enabling efficient management of resources and real-time performance in virtualized environments.

1. RT-Xen: Real-Time Virtualization
Sisu Xi, Meng Xu, Chenyang Lu, Linh T.X. Phan, Christopher D. Gill, Insup Lee, Oleg Sokolsky

2. Real-Time Virtualization
Cars: Consolidate ~100 ECUs -> ~10 multicore processors
Infotainment on Linux or Android
Safety-critical control on AUTOSAR
Cloud Computing’s Killer App: Gaming [IEEE Spectrum]
Need to compute and stream 30 to 50 frames per second
1
Applications must meet real-time performance constraints on virtualized platforms!

3. RT-Xen: Real-Time Virtualization
Real-time hypervisor scheduling framework in Xen
Implement a suite of real-time scheduling algorithms
Based on compositional scheduling theory
VMs specify resource interfaces
Real-time guarantees to tasks in VMs
Open source
RT-Xen patch submitted
https://sites.google.com/site/realtimexen/
2

4. Xen Virtualization Architecture
Guest OS runs on VCPUs
Hypervisor schedules VCPUs on PCPUs
Credit scheduler
[Weight, Cap] per VM
Round robin
3

5. RT-Xen Interface
VM resource interface
A set of VCPUs, each characterized by <period, budget>
Optional: use cpumask to specify VCPU affinity with PCPUs
Hide task-specific information
Real-Time scheduling algorithms
Ordering of VCPUs? Priority scheme
Placement of VCPUs?Global vs. partition
Resource isolation?Server mechanisms
4

6. Real-Time Scheduling Policies
Priority scheme
Static priority: Deadline Monotonic (DM)
Dynamic priority: Earliest Deadline First (EDF)
Global scheduling
Schedule VCPUs based on global information
Allow VCPU migration across cores
Flexible use of multiple cores
Migration overhead and cache penalty
Partitioned scheduling
Assign and bind VCPUs to PCPUs
Schedule VCPUs on each core independently
May underutilize PCPUs
No migration overhead or associated cache penalty
5

7. Scheduling a VCPU as a Deferrable Server
6
time
T1 (10, 3)
T2 (10, 3)
0
5
10
15
0
5
10
15
time
Deferrable Server
(5,3)
Budget
A VCPU receives budgetus of CPU resources every periodus
Budget is replenished at every start of period
VCPU consumes budget when running, suspends when no budget left
Preserves budget when there is no task

8. RT-Xen Investigation Roadmap
Single-coreRT-Xen 1.0
Single-core enhancedRT-Xen 1.1
Multi-coreRT-Xen 2.0
7
Global
Scheduling
Fixed Priority
(DM)
Partitioned Scheduling
Dynamic Priority
(EDF)
Periodic
Deferrable
Periodic
Deferrable
Deferrable
Periodic
Deferrable
Periodic
Sporadic
Polling
Work Conserving Periodic
Capacity Reclaiming Periodic
gDM
gEDF
pEDF
pDM

9. RT-Xen 2.0: Run Queues
A run queue
holds VCPUs that are runnable (have task to run)
has two parts: VCPUs with budget and out of budget
is sorted by priority (DM or EDF) within each part
rt-global: all cores share one run queue with a spinlock
rt-partition: one run queue per core
Patches for more efficient implementation on the way!
8
RunQ
VCPUs withbudget
VCPUs out of budget
Sorted by priority

10. Experimental Setup
Hardware: Intel i7 processor, six cores running at 3.33 GHz
Dedicate one PCPU to domain 0
All guest VMs use the remaining cores
Cache architecture
Each core has dedicated L1 cache (32 KB) and L2 cache (256 KB)
All six cores share L3 cache (12 MB)
Inclusive L3 cache, all data in L2 cache must also be in L3 cache
Software
Xen 4.3 patched with RT-Xen
Guest OS: Linux patched with LITMUSRT
9

11. RT-Xen 2.0: Scheduling Overhead
10
rt-global has extra overhead due to global lock
credit has high max overhead due to load balancing

12. RT-Xen 2.0: Credit Scheduler
11
credit missed deadline at 22% CPU capacity
RT-Xen delivers real-time performance up to 78%

13. RT-Xen 2.0: Real-Time Performance
12
Global scheduling wins empirically!
gEDF + deferrable server -> best real-time performance

14. Demo
YouTube: “RT-Xen Demonstration”
https://www.youtube.com/watch?v=wisxWn3mR5s
13

15. Patch Status
Patch RFC v1 & v2July 10th& July 29th
gEDF + deferrable server
cpupool support
Patch RFC v3Expected on Aug 24th
Scheduling trace support
Performance improvement (splitting RunQ)
Patch RFC v4Expected before Sep 10th
Performance improvement
•Timer based budget replenishment
•Improve the timing resolution of budget
14

16. Conclusion
Diverse applications demand real-time virtualization
Real-time virtualization in embedded area
Cloud gaming
RT-Xen provides real-time performance
Efficient implementation of diverse real-time scheduling policies
Leverage compositional scheduling theory -> analytical guarantee
gEDF + deferrable server wins empirically
15

17. Research Contributions
RT-Xen 1.0:S. Xi, J. Wilson, C. Lu, and C.D. Gill, RT-Xen: Towards Real-Time Hypervisor Scheduling in Xen, ACM International
Conferences on Embedded Software (EMSOFT) 2011
RT-Xen 1.1: J. Lee, S. Xi, S. Chen, L.T.X. Phan, C. Gill, I. Lee, C. Lu and O. Sokolsky
Realizing Compositional Scheduling through Virtualization, IEEE Real-Time and
Embedded Technology and Applications Symposium (RTAS) 2012
RT-Xen 2.0:S. Xi, M. Xu, C. Lu, L.T.X. Phan, C. Gill, I. Lee, and O. SokolskyReal-Time Multi-Core Virtual Machine Scheduling in Xen, ACM International
Conferences on Embedded Software (EMSOFT) 2014
RT-Xen 2.1 + RT-OpenStack: S. Xi, C. Li, C. Lu, C. Gill, M. Xu, L.T.X. Phan, C. Gill, I. Lee, and O. SokolskyRT-OpenStack: Co-Hosting RT VM with non RT VMs, in submission
RTCA:S. Xi, C. Li, C. Lu, and C. GillPrioritizing Local Inter-Domain Communication in Xen, ACM/IEEE International
Symposium on Quality of Service (IWQoS) 2013
RT-Xen patch (gEDF with deferrable server)
RT-Xen: Real-Time Virtualization in Xen, Xen Blog, 2013
RT-Xen: Real-Time Virtualization in Xen, Xen Developer Summit, 2014
16

18. Backup Slides
17

19. RT-Xen 1.0
18

20. Implementation –PCPU 19
Budget
XBudget
Task
RunQ
RunQ
XTask
RdyQ
RdyQ
VCPU Position
Three Queues within One Physical Core
VCPU Params
(period, budget, priority)
IDLE
Periodic
０
３
IDLE

21. Server Design –Deferrable & Polling
Servers (Period, Budget, Priority)
S1 (5, 3) with Two Tasks
T1 (１０, ３ )
20
T2 (１０, ３ )
Time
5
０
１０
１５
Budget in S1
Actual Execution
Deferrable Server
Time
5
０
１０
１５
３
back-to-back
Budget in S1
Actual Execution
Polling Server
Time
5
０
１０
１５
３
1.Replenish?
2.Budget but NO task?

22. Server Design –Periodic & Sporadic
Servers (Period, Budget, Priority)
S1 (5, 3) with Two Tasks
T1 (１０, ３ )
21
T2 (１０, ３ )
Time
5
０
１０
１５
Budget in S1
Actual Execution
Sporadic Server
Time
5
０
１０
１５
３
5
+3
5
+3
overhead ++
Budget in S1
Actual Execution
Periodic Server
Time
5
０
１０
１５
３
theory favored
1.Replenish?
2.Budget but NO task?
?

23. RT-Xen 2.0
22

24. RT-Xen 2.0: Workload
Periodic task sets: [period, execution time, deadline]
CPU-intensive, independent tasks
Randomly generate the task sets until a total task utilization, then distribute tasks to four VMs, and apply compositional scheduling theory to calculate each VM’s resource interface
25 task sets per data point, measure fraction of schedulable tasksets
23

25. RT-Xen 2.0: Context Saved
8/20/2014 24
Less than 1 us overhead for the spinlock

26. RT-Xen 2.0: Theory vs. Experiments
25
gEDF < pEDF theoretically due to pessimistic analysis
gEDF > pEDF empirically, thanks to global scheduling

27. RT-Xen 2.0: Theory vs. Experiments
8/20/2014 26
gEDF > pEDF empirically, thanks to global scheduling
gEDF < pEDF theoretically due to pessimistic analysis

28. RT-Xen 2.0: How about Cache?
27
Benefit of global scheduling dominate migration cost on a shared L3-cache platform.

29. RT-Xen 2.0: Context Switch
8/20/2014 28
“Fake” context switch with idle VCPUs