Virtual Mobile Cloud Network for Realizing
Scalable, Real-Time Cyber Physical Systems�
Kiran Nagaraja, Yanyong Zhang,
Ivan Seskar, Dipankar Raychaudhuri (PI)
WINLAB, Rutgers University
Kiyohide Nakauchi, Yozo Shoji (PI)
NICT
JUNO PI Meeting
June 25, 2014�
Summary
n
Design & Develop
Virtual Mobile Cloud Network (vMCN)
Ø
Trillion-order scalability
for CPS devices/objects
Ø
Less than
100 msec response time
in CPS application
n
WINLAB’s prior works mainly for the scalability are
organically integrated with NICT’s prior works mainly for
the real-time
n
Additional research challenges related to
cloud migration
and
virtual network design
will be addressed jointly by NICT
and WINLAB
n
Demonstrate vMCN with a typical CPS application over GENI
Needs for Real-time & Scalable Cyber Physical
System (CPS) are Emerging�
Emergency stop
…
Machinery Control
Emerging Services
AR navigation
Automatic translation
Automatic driving
Emergency alert
Safety
Mobile Cloud Services
Scalability for
devices and objects
Real-time
Object GUID�� Value 1� X1� 2� X2� …� …� 10^12� Object GUID�� Value 1� X1� 2� X2� …� …� 10^12�
Future CPS Function Model and Requirements
Cloud
Wearable� Smart phone�Camera sensor�
(1) Sensing�
(2) Computation�
(3) Actuation�
Response�
Object ID�� Value 1� X1� 2� X2� …� …� 10^12�Query�
Machine�- Device/Object ID
- Captured/Sensed data
�
- Information
- Actuation command�
Tag/RFID� Contents�Large computational
workload �
Large # of
objects/contents�
Network domain . . .�
Large # of
NW domains�
Large # of
CPS devices�
<100msec
response time in
application
＆
＆��
Trillion-order
are properly handled
CPS devices & objects
n
Scalable
��
n
Real-time
��
n
Terminal Mobility��
Service
Directories
Project Goals
n
Design a
network architecture
Ø
Trillion-order scalability for CPS devices/objects
Ø
Less than 100 msec response time in CPS application
n
Develop a
proof-of-concept prototype
Ø
MF router software enhanced for the real-time
Ø
vBS software enhanced for global mobility
Ø
Running the prototype over GENI and JGN-X
n
Demonstrate an example future
CPS application
Ø
AR (Augmented Reality) application using glass
Overview of Virtual Mobile Cloud Network (vMCN)
�
n
At-scale & low-latency global name resolution
n
Dynamically configurable wired and wireless resources
n
Optimal placement and dynamic migration of cloud services
ü
Scalable�
ü
Real-time�
ü
Real-time�
IDQuery Packet
(2) VN & vBS on
the move
(3) Dynamic Cloud
Migration
Scalability
Real-time
response
Routing layer
Name resolution layer
Virtual base station (vBS)
CPS slice
(1) Name-based
network
Technical Approach :
(1) Name-based Network
Hybrid GUID/NA
based Forwarding
GNRS
DATA SType NAs GUID NAsFlexible Resolution
1. Early binding
2. Late binding
3. Dynamic binding
Internet scale
evaluation
(26K networks):
cost of resolution
< 100ms
Spatial locality
awareness can
reduce it further:
few 10s of ms
n
Key MobilityFirst components
n
At-scale, low-latency
global name resolution
n
Flexible and extensible
hybrid GUID/NA routing
n
Research issues
n
Exploiting locality to minimize name resolution costs and RTTs
n
Spatial locality-aware GNRS
Technical Approach :
(2-1) Virtual Network Support in MobilityFirst
n
Service-specific Virtual Network (VN)
n
GUID can be used to name end-to-end service slice
n
VN required to span network elements and end hosts
n
Extensions required for both name resolution and routing layer
GNRS mappings
can capture VN
membership and
access control
Technical Approach :
(2-2) Virtual Base Station (vBS) on the move
Roaming
virtual Base
Station (
vBS
)
for CPS
User devices
Logically integrated BS
with
extensible wireless
capacity
�
Managed association and
seamless handover
�
Physical Base
Stations
CPS Slice
n
Key vBS components
n
Dynamically provisioning physical BS resources
for target applications
n
Adaptation to mobility
in a “local” domain
n
Research issue
n
When and where physical BS resource should be provisioned
in global
Technical Approach :
(3) Dynamic Cloud Migration
n
Key components
n
Service placement and migration for quick response in CPS application
n
Routing adaptation and service continuity
n
Research issues
n
Optimization algorithms for VM placement and migration
n
Anycast routing approaches for dynamic cloud services
CPS Slice
Core
Cloud
Detecting mobility by
monitoring RSSI
Mobility-driven
Service Migration (Localization)
Send (GUID = 11011..011, SID=anycast, data)
DATA SID GUID
NA99
NA32
Service anycast
ServiceID
Mobility
Edge
Cloud
Why and How We Collaborate?:
Specific Design Issues and Research Questions
vMCN
Low-latency global name
resolution
Mobility-driven
dynamic cloud migration
Virtual name-based
network
“Global” vBS on the move
Service anycast
MobilityFirst
(WINLAB)
vBS (NICT)
Our Prior Work�
Collaboration�
Global name resolution
GUID-based routing
Cloud resource localization
vBS on the move
BS virtualization
Hybrid GUID/NA routing
Storage-aware routing
Hop-by-hop transport
How to reduce lookup
latency in a
global-scale DHT?
When & where VMs
should be migrated?
How to realize service
ID based anycast on
GUID-based networks?
How to realize virtual
network capability on
GUID-based networks?
When and where
physical BS resource
should be provisioned?
Research Questions�
Integrate
Integrate
Seamless handover
Integrate
Final Demonstration Image
n
(Performance Goal) response time is less than 100msec for 10^6 queries/sec
n
(Function Goal) CPS slices can be built over US-JP wide-area MobilityFirst network
n
(Application Goal) AR service w/ glass devices over a proof-of-concept prototype
WiFi�
WiMAXWiFi
WiMAX BTS
CPS
Server
CPS
Server
CPS App
MF
CPS App
MF
MF
CPS
Server
Dynamic Cloud
Migration
vBS on the move
MF
MF Routers
. . .� . . .�vBS
. . . .�(Emulated) CPS devices
JGN-X
GENI
Virtual Name-based
Network
ORBIT
Summary
n
Design & Develop
Virtual Mobile Cloud Network (vMCN)
Ø
Trillion-order scalability
for CPS devices/objects
Ø
Less than
100 msec response time
in CPS application
n
WINLAB’s prior works mainly for the scalability are
organically integrated with NICT’s prior works mainly for
the real-time
n
Additional research challenges related to
cloud migration
and
virtual network design
will be addressed jointly by NICT
and WINLAB
n
Demonstrate vMCN with a typical CPS application over GENI
Milestones�
NICT
Rutgers
NICT
Deliverables�
Project Task Responsible
Groups
Year 1 Deliverables Year 2 Deliverables Year 3 Deliverables (6 months)
T.1. MF-‐based virtual network architecture & protocol implementa�on
WINLAB Design and preliminary
evalua�on; so�ware prototype; tes�ng on ORBIT SDN sandbox
MF/VN so�ware
implementa�on and integra�on with MF Click & OpenFlow s/w
Version update T2. MF op�miza�ons for
cloud service including GNRS speed-‐up & cloud migra�on
WINLAB Op�miza�on of MF services (anycast, context, compute) for cloud service and GNRS speed-‐up via caching etc.
Upgrade to MF code release including service API; GNRS extensions and integra�on
-‐-‐-‐-‐
T.3. Local vBS design enhancements and latency reduc�on
NICT Design and evalua�on of state transfer method for local VN
implementa�on of vBS migra�on protocols and algorithm in BYON prototype
-‐-‐-‐-‐
T4. BYON-‐based virtual network design and implementa�on
NICT Design and preliminary
evalua�on; BYON-‐API for vBS and cloud migra�on; tes�ng on BYON prototype
vBS so�ware implementa�on and integra�on with BYON OpenFlow base sta�on
Version update
T.5. vMCN/Global vBS on the move design,
integrated with MF GUID services
NICT and WINLAB
Integrated design of BYON/vBS and MF/VN; Evalua�on of GUID-‐ based method for global VN
Integra�on of vBS’s code with MF protocol stack; global area cloud migra�on prototype
Version update T.6. Proof-‐of-‐concept vMCN system prototyping & applica�on demos NICT and WINLAB Preparing/leveraging key
components: MF protocol stack, GNRS, VN extensions, vBS extensions, AR app, etc.
Integra�on of US-‐Japan prototype on GENI and JGN; applica�on tes�ng; system performance studies
Integrated proof-‐of-‐concept demos& CPS applica�on trials
