Transcript No Slide Title

Research Overview:
What Sayeem Has Been Doing?
Abu (Sayeem) Reaz
University of California, Davis, USA
National Instruments Interview
February 09, 2011
Page 1
Earliest Multi-Hop Network
Betterment of networks using
feasible technologies
Andreas J. Kassler, Research Opportunities at Karlstads Universitet
Page 2
Presentation Overview
• PhD Research
• Routing over Wireless and Optical Access
• Asymmetric “Capacity” Deployment and Resource Assignment
• Integrating Cloud in Access Network and Green Routing
• Wireless Highway for 3G Backhaul
• IPTV Stream Generator
• MS Research
• Location Management using DNS
• Multi-class (Vertical) Handoff Management
• Secure Paging in Handoff Management
• Opportunity for Contribution to NI
• Problem Solving
• Programming and Development
Page 3
PhD Research
Page 4
Network Architecture: WOBAN (1)
Page 5
Network Architecture: WOBAN (2)
•
WOBAN: Wireless-Optical Broadband Access Network
•
Deploy broadband access network with minimum wiring: cost effective
•
An optimal combination of optical and wireless network to minimize cost and maximize
utilization and performance
•
•
Back-end: Optical access network, e.g., Passive Optical Network (PON)
•
Front-end: Multi-hop Wireless Mesh Network (WMN)
Optical Scenario:
1. Optical Line Terminals (OLTs) at Central Office (CO) are connected to Optical
Network Units (ONUs) via fiber
2. ONUs are connected to the wireless access network via gateways
•
Wireless Scenario:
1. A set of wireless routers form a wireless mesh network: end users are connected to
nearby router
2. Some wireless routers work as gateways, connecting the wireless network to optical
network
Page 6
Why?
WMN
+
PON
We like to have our cake and eat it too!
Page 7
Routing: The Big Picture
Efficient routing across WMN and PON: Shortest Delay
Page 8
WMN: Divide the Capacity
Asymmetric
Page 9
PON: Native Routing
Downstream:
Broadcast
Upstream:
Dynamic Bandwidth
Allocation
Page 10
Data Flow
Downstream
Upstream
Page 11
Summary
Page 12
Asymmetry in WOBAN
Traffic flows to and from the OLT
Bottleneck near the Gateways
Flow Aggregation
Page 13
As a Result…
Many “links” are not even used!
Not all nodes need the same Capacity
Traffic on Links (Mbps)
Page 14
Mixed Capacity Wireless Access
Deploy radio where needed!
Page 15
Radio Deployment: MILP
Page 16
Summary
Page 17
Resource Assignment: Challenges
Asymmetric Capacity and Flow
Need to assign both Radio and Channel
Page 18
Traffic Aggregation
Smoother instantaneous burstiness!
http://www.ams-ix.net/technical/stats/
Page 19
Channel Assignment: BLP
Intelligent Channel and Radio Assignment (ICRA)
Page 20
Summary
Page 21
Bringing Service to Users
Service = Content and/or Application
Can we bring them to closer to users?
Cloud-Integrated WOBAN (CIW)
Alix Boards
Clougplug
Page 22
Service Access: Traditional
Page 23
Service Access: CIW
Page 24
What Can We Gain
• Adds value to the network  Competitive Edge
• “Now I want to use this network!!”
• Remove device dependencies
• Any common interface: possibly a browser
• Local services requests are delivered locally
• No/Limited traffic introduced to wireless backhaul
• More room for regular mesh traffic
• Service traffic moves away from gateways
• Bottleneck reduced
• Local updates remains local
• Likelihood of stale information becomes low
Page 25
Implementations
Wisper
Firetide
Aruba/Tropos/Meraki
Page 26
Deployment of CC: MILP
Page 27
Summary
Page 28
Green Routing in CIW (GRC)
Different part of
the network is
busy at different
time of the day
Page 29
GRC
Instead of pack-and-turnoff, utilize the architecture of WOBAN:
Selective Turnoff and Load Balance
3. Load balance
for each pipe
1. Split into Zones
2. Create BW Pipe
for each Zone
Page 30
Path Computation: Auxiliary Graph
Page 31
Summary
Page 32
3G Backhaul
AT&T’s 3G cell sites are backhauled
primarily through T1 lines, which, while
adequate in the early days of UMTS, wind
up becoming a choke point as AT&T
upgrades to faster and faster network
technologies.
Connected Planet, Jan, 2010, http://connectedplanetonline.com/3g4g/news/att-doubles-3g-010510/
Page 33
3G Architecture
Is fiber capacity properly utilized?
Is copper a bottleneck?
Single point of failure?
Page 34
Without Huge Investment…
Can we develop a methodology to
• utilize fiber capacity
• reduce copper bottleneck
• create alternate paths for failure recovery
• provide better service quality to high bandwidth
application
- Broadcast TV to UE
An Overlay Network adjunct to the existing 3G
network using High Capacity Wireless Links
Page 35
Overlay Network Architecture
Links become backup
of each other
P2P High Capacity
Wireless Link
Load Sharing
Page 36
The Big Picture
Multiple Overlays
Any size,
any shape
Page 37
Overlay Placement: MILP
Page 38
Summary
Page 39
The WMN Version of the Problem
We have also investigated how an Overlay Network
can be deployed in WMN
Because of the interference within the WMN, this is actually
a “harder” problem
Page 40
and the Formulation without the Details…
Page 41
Summary
A 43-Node WMN with 3 Gateways
Tested for deployment of 1, 2, and 3 overlay links
Page 42
I and B Frame from Trace
Correlated yet Different!
Page 43
I and B Frame: Distribution
We need to generate I and B frames separately
Lognormal distribution closely approximates the frame size distribution of
I and B frames
M. Krunz and H. Hughes, “A traffic model for MPEG-coded VBR streams,'' Proc.,
ACM SIGMETRICS, 1995.
Page 44
New Scene
Videos are constructed with scenes!
Scene length is important:
Within a scene, I frame sizes are close to each other…
Ik
∆
Ik+1
If ∆ is significant, then it’s a new scene!
M. Krunz and H. Hughes, “A traffic model for MPEG-coded VBR streams,'' Proc.,
ACM SIGMETRICS, 1995.
Page 45
Scene Length Distribution
Page 46
Variation Within a Scene
We use the relative sizes of all
the I frames in a scene
compared to the first I frame
Addresses the variations within a scene
Page 47
Data Rate on 10G EPON
Each frame size was picked from corresponding Lognormal
distribution, but relation between scenes is not considered
Increased and continuous burstiness
Page 48
Relative I Frame Size
We use the relative sizes of the first
I frame in every scene and generate
subsequent I frame sizes in the
scene from the first I frame size
Page 49
Relative B Frame Size
We use the relative B frame
sizes compared to the I
frame size in a GoP
Page 50
Resultant Synthetic Trace
Correlated, spike free synthetic traces with proper variations
Page 51
Distribution of Frame Sizes
The frame size distributions match targeted Lognormal distributions even
though they are not generated from actual Lognormal distributions
Page 52
Original vs. Synthetic Trace
Voila!
Page 53
Data Rate on 10G EPON
Page 54
MS Research
Page 55
IP Mobility
Old point of attachment
Subnet 1
New point of attachment
IP Address 1
IP Address 2
(old location)
(new location)
Subnet 2
Page 56
SIGMA: Seamless IP-diversity based
Generalized Mobility Architecture
Basic idea: setup a new path to
communicate with CN while
maintaining the old path.
Handover process:





STEP 1: Layer 2 handover
and obtain new IP address
STEP 2: Add IP addresses
into the association
STEP 3: Redirect data
packets to new IP address
STEP 4: Update location
manager (LM)
STEP 5: Delete or deactivate
obsolete IP address
Step 3
1
2
4
5
CN
LM
Internet
Router
MH
Subnet 2
Subnet 1
1 IP Address
2 IP Addresses
1 IP Address
Page 57
Location Management using DNS
2
Location Update
DNS
Internet
3
1
Subnet 1
IP Address 1
IP Address 2
Subnet 2
CN
Page 58
Challenge
Failure
Query time > Duration in Overlapping Area
Page 59
Mobility Model
Determine if there will be a query to DNS while updating the entry
Page 60
Summary
Page 61
Multi-Class Handoff: mSIGMA
Page 62
Handoff Decision
Page 63
Performance
Page 64
Packet Trace
WLAN
to
CDMA
CDMA
to
WLAN
Page 65
Paging SIGMA: P-SIGMA
•
PA  single ID for subnets
•
Roam within PA without updating LM
•
Active and Idle MHs update DNS at
inter-PA handoff
•
Active MH updates PGW at intra-PA
handoff
DNS
PGW 1
PGW 2
Internet
Location Update
Idle MH
Active MH
Active MH
ID = X
ID = X
ID = X
Subnet 1
Subnet 2
Subnet 3
= Y PA,
HandoffID
across
Location Updates
Handoff within PA, Handoff within PA,
Update PGW and DNS
Update PGW
No update
PA 1
Subnet 4
PA 2
Page 66
Paging Algorithm
•
Low mobility  last location paging
•
High mobility  fixed paging
DNS
PGW 1
PGW 2
Internet
Low mobility subnet
MH not found
High mobility subnets
MH found
MH found
paging with MAC2
paging with MAC1
paging with MAC2
paging with MAC1
Last location paging
Subnet 1
Subnet 2
PA 1
Fixed paging
Subnet 3
Subnet 4
PA 2
Page 67
Connection Initiation
•
•
PGW is lightweight LM
 updated for only
active hosts
DNS is heavyweight
LM  updated for all
hosts only for inter-PA
handoff
CN
DNS
name lookup
IP address X
Location Update
PGW 1
PGW 2
Internet
Connection INIT
Low mobility subnet
High mobility subnets
Registration with
IP address Y
Connection INIT to Y
IP address X
IP address Y
MH with MAC A
paging for A
Subnet 1
Subnet 2
PA 1
paging for A
Subnet 3
Subnet 4
PA 2
Page 68
Attack on P-SIGMA
Page 69
Free Loading
Session Hijacking
Intrusion Detection Algorithm
Page 70
Summary
Page 71
Opportunity for Contribution to NI
Page 72
Problem Solving
• Identify new challenges for NI products
• Using optimization techniques to maximize performance
• Linear Programming
• Simulated Annealing
• Apply networking techniques
• For intelligent data-flow
• Energy efficiency
Page 73
Footstep on a New Area
In a nutshell, lightRadio takes all of the essential
elements of traditional base stations and antennas and
shrinks them so that they can be distributed across the
access network -- or cloud -- and deployed dynamically
where or when capacity and coverage is needed. And
the distributed network elements are connected via
fiber-optic networks.
http://www.lightreading.com/document.asp?doc_id=204081
Page 74
Programming and Development
• Development of network-related products
• Design intelligent protocols for routing
• Implement upper-layer protocols using socket
programming
• Implement stack for lower-layer protocols
• Use generic programming skill to contribute to any
development
Page 75
Network Programming
• For Network layer or higher
• Use native TCP/Datagram socket
• For MAC layer
• Raw socket programming for common MAC protocols
• Send and receive data using MAC address
• IRQ to access registers
• Extract information from driver (not familiar)
Page 76
Thank you!
Contact Information:
E-mail: [email protected]
Phone: 530-574-2090
Web: http://networks.cs.ucdavis.edu/~sayeem/
Page 77