The document describes research into deploying virtualized e-learning applications with real-time guarantees on the IRMOS cloud computing platform. It discusses modeling network and computing delays, benchmarking an e-learning application in a virtualized environment, and using neural networks to predict performance metrics like response time. The goal is to enable precise quality of service for interactive real-time applications in the cloud.
WordPress Websites for Engineers: Elevate Your Brand
Virtualised e-Learning with Real-Time Guarantees on the IRMOS Platform
1. IEEE 2010
December 13 – 15, 2010
Perth, Australia
Virtualised e-Learning with Real-Time
Guarantees on the IRMOS Platform
Tommaso Cucinotta
Real-Time Systems Laboratory
Scuola Superiore Sant'Anna
Pisa, Italy
… and 12 others from 6 institutions:
3. Introduction
Towards a new computing paradigm
Computing, network, storage in the cloud
Not only batch computing and storage
but also interactive real-time applications
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
4. Web and Clouds
Performance Today
How much should I replicate my
infrastructure
to meet desired
average QoS
levels ?
General-purpose
technologies
What about
the non-average
cases and interactivity?
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
5. The IRMOS Approach
to real-time and stable QoS
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
7. IRMOS
Two-Phases Approach
Design
Tools
Benchmarking Application
Concretion Discovery
Negotiation
Modeling, Reservation
Mechanisms for
Mechanisms for
Methodology for the
Analysis,
Methodology for the
Planning precise allocationService
precise allocationof of
identification of
identification of resources
resources Instantiation
resource requirements
resource requirements to applications Service
to applications
Component
Configuration
Execution &
Monitoring
Cleanup
Offline 7
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
13. Real-time e-Learning
Synchronising real and virtual
worlds
Venue locations
Access devices
Interactive
Gallery media and communication
terminals Real time synchronization
Mobiles
Festivals
Interactive Virtual Worlds
whiteboards
Learning
Media generation
Remote locations
Professional
Museums
Social
Networking
Personal Classroom
Home
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
14. Mobile e-Learning
Architecture and model
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
15. Mobile e-Learning
Architecture and model
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
16. Deploying e-Learning
Goal Scheduling params
(budget, period)
Deploy the e-Learning server
With the application-level parameters
specified in the SLA
Detail level
(i.e., resolution)
Maximum number
of users
Respecting the SLA QoS Physical Host
Statistics on the
response-time of
individual requests mean Resol.
Resol. W x H x fps
W x H x fps
(mean, max, std dev) std-dev Users
Users 10
10
max
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
17. Modelling e-Learning
Non-IRMOS world
Network performance highly dependent
on traffic of other apps
Computing performance highly
dependent on workload of other apps
When deployed in IRMOS/ISONI
QoS-aware networking and CPU real-time
scheduling limit the interferences
among different application instances
Applications can be analysed in isolation
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
18. Modelling e-Learning
Network delays: Erlang distributions
Parameters fitted on benchmark data
Computing delays
Strong dependence on application-level
parameters (number of users, resolution)
Black-box approach → Neural Networks
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
19. Benchmarking Data
The real-time
scheduler
successfully
isolates
performance
of 2 VMUs
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
21. Neural Network
Training
After training, the
ANN successfully
outputs the mean
and standard
deviation of the
SC response time:
prediction error
less than 3%
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
22. Conclusions and future work
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
23. Conclusions
The IRMOS/ISONI virtualized
infrastructure facilitates
benchmarking & modelling
off-line performance prediction
on-line performance stability
allowsfor better server consolidation levels
while meeting the timing constraints
We showed the IRMOS way to deploy an
e-Learning application with precise
QoS guarantees
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
24. Future work
(WiP, actually)
Apply the methodology to the
VirtualWorld e-Learning platform
Apply the methodology to the other
IRMOS application scenarios
Film post-production
Virtual reality in automotive
Model how the scheduler affects the
QoS metrics, to reduce the number
of configurations to benchmark
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
25. References
T. Cucinotta, K. Konstanteli, T. Varvarigou, "Advance Reservations for Distributed
Real-TimeWorkflows with Probabilistic Service Guarantees", IEEE International
Conference on Service-Oriented Computing and Applications (SOCA 2009),
December 2009, Taipei, Taiwan
K. Kostanteli, D. Kyriazis, T. Varvarigou, T. Cucinotta, G. Anastasi, "Real-time
guarantees in flexible advance reservations", 2nd IEEE International Workshop on
Real-Time Service-Oriented Architecture and Applications (RTSOAA 2009),
Seattle, Washington, July 2009
F. Checconi, T. Cucinotta, D. Faggioli, G. Lipari, "Hierarchical Multiprocessor CPU
Reservations for the Linux Kernel", in 5th International Workshop on Operating
Systems Platforms for Embedded Real-Time Applications (OSPERT 2009), Dublin,
Ireland, June 2009
T. Cucinotta, G. Anastasi, L. Abeni, "Real-Time Virtual Machines", in 29th Real-
Time System Symposium (RTSS 2008) -- Work in Progress Session, Barcelona,
December 2008
YouTube Video on e-Learning performance isolation:
http://www.youtube.com/watch?v=8FbHZ4ytNoQ
IRMOS YouTube channel:
http://www.youtube.com/user/irmosproject
IRMOS Project Website: http://www.irmosproject.eu
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it
26. Thanks for your attention
Questions ?
Tommaso Cucinotta – Real-Time Systems Lab (RETIS), Pisa, Italy – cucinotta@sssup.it