Transcript MSODA - Purdue University

ACM Multimedia 2004
Towards an Integrated Multimedia Service
Hosting Overlay
Dongyan Xu, Xuxian Jiang
Department of Computer Sciences
Center for Education and Research in Information
Assurance and Security (CERIAS)
Purdue University
Outline






Motivation
MSODA architecture
MSODA components
Virtualization of service hosting overlay
Related work
Conclusions
Motivation
 Proliferation of value-added and function-rich
media services
 Pervasive media sources: live cam, TV, radio…
 Content-based processing: tracking, enhancement,
mix-reality…
 User-specific media service composition:
 Surveillance cams  image recognition  scene
correlation
 Home video  jitter elimination  music mixing 
mixed-reality rendering
Image Repair Summarization Music Mixing
Motivation
 Service oriented architectures
 Users don’t have to know
 Service implementation details
 Service instance locations
 Service-level routing decisions
 Service providers have more flexibility in
 Implementation
 Deployment strategy: placement, replication, migration,
resource scaling, coalition
 Management: upgrade, troubleshooting, recovery
Motivation
 Service providers meet service host
Service providers:
Have no infrastructure
For deployment
Service host (e.g. Yahoo, MSN):
Needs rich services
to serve customers
A service-oriented “marketplace”:
Hosts a large variety of media services
for customer access and composition
Challenges
 Decoupling service management from hosting
platform management
 Isolating management of different media services
 Protecting hosting platform from untrusted media
services
 Enabling agile media service workflow optimization
 On-demand service capacity scaling
 Service instance replication and re-location
Our Solution: MSODA
(Media Service On-Demand Architecture)
 Infrastructure: MSODA hosts in wide-area network
 Media service instances : virtual machines in
MSODA hosts
 Media service cloud : virtual network of service
instances
 Service gateways : edges of service cloud and
interface to customers
MSODA Architecture
Service
Instance
(VM)
MSODA
host
Service
gateway
MSODA Host
 Two-level architecture
 Host
 Virtual machines
 Host domain MSODA
daemons




Resource allocation
Network monitoring
Traffic tunneling
Service routing
S1
S2
…
…
Guest OS Guest OS
…
MSODA daemons
Host OS
An MSODA host
MSODA Gateway
 Interface to service clients
 Service composition
 Service configuration
 Edge of service cloud
 Bridging service instances (virtual machines) to
client machines: limited and controlled access
Composite
service request
Client
Service path signaling
Service data/stream
MSODA
gateway
Service
instance (VM)
MSODA Gateway
 Service composition and configuration
 User-centric customization
 Resource conservation
S1
S2
512Kbps
S2
S1
S2
Media Service Cloud
 A virtual network of service instances (VMs)
 Based on network virtualization technique (VIOLIN)
 VN for VMs
 Using MSODA hosts as underlying carrier (layer-2 on UDP)
 Emulating advanced network protocols (e.g., IP multicast)
 IP-compliant, with its IP address space
 Isolation from underlying Internet
Media Service Cloud
 Advantages
 Protection of MSODA infrastructure
 Service traffic volume control
 Service instance reachability control
 Decoupling of
 Media service function (by service developer)
 Service provisioning and composition mechanisms (by
MSODA developer)
Media Service Cloud
 Multicast and anycast group for each media
service
 Multicast group: convenient service management
(e.g., asking all instances of a service to report
current load/QoS/most popular content…)
 Anycast group: service composition routing (e.g.,
specifying the next service in the service delivery
workflow)
 Simple APIs for easy media service implementation
 Actual operations performed by underlying MSODA
hosts
Media Service Cloud
 Dynamic service cloud evolution
 Service instance resource scaling
 Service instance replication
 Service instance re-location
S1
S1
S1
S2
S2
Resource
scaling
Time
Service
instance
replication
S2
MSODA Prototype
 Service instances (VMs) enabled by User-Mode
Linux (UML)
 Service cloud (virtual network) enabled by VIOLIN
 Acceptable network performance degradation
 Automatic service instance creation and re-location
 Centralized computation of service delivery paths
 Local and wide-area (PlanetLab-based) testbeds
 Virtual private Grids for dynamic scientific
applications
Related Work
 Service composition frameworks
 Ninja, SAHARA, CANS, SPY-Net, SpiderNet
 Service overlay networks
 SOI (Service-Oriented Internet)
 Opus (Overlay Peer Utility Service)
 Overlay networking
 RON, OverQoS, Narada, Overcast, I3
 Resource virtualization
 Virtual machine: Denali, VMware, UML, Xen
 Virtual network: VNET, VIOLIN
 Virtual environment: In-VIGO
Conclusions
 MSODA: an integrated media service hosting
platform for service composition
 Virtual machine as granularity for service
instance management and manipulation
 Virtual service cloud network
 Platform-independent media service
development and management
 Maximum manipulability for dynamic service
instance scaling, replication, and re-location
 Strong protection of MSODA platform from
untrusted media services/clients
Thank you.
For more information:
Email: {dxu, jiangx}@cs.purdue.edu
URL: www.cs.purdue.edu/~dxu
Google: “Purdue SODA friends”