Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

European Wireless 2018 Keynote

No description
by

Zhangyu Guan

on 24 September 2018

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of European Wireless 2018 Keynote

Considerations
Figure from
Nick McKeown, et al., "OpenFlow: Enabling Innovation in Campus Networks," ACM SIGCOMM Computer Communication Review, vol. 38, no. 2, Pages 69-74, April 2008.
Data
Plane
Prof. Tommaso Melodia

Department of Electrical & Computer Engineering
Northeastern University

Decomposition Framework
Network:
Generalized Network Utility Maximization (NUM)

Layering architecture:
Decomposition scheme

Layers:
Decomposed subproblems

Interfaces:
Functions of primal and dual variables

Programmable Protocol Stack
Architectural framework to ease the implementation of cross-layer protocols
Provide abstraction of a radio protocol stack and building blocks

Retain control of implementation details at all layers

Maintain platform independence

Given distributed optimization algorithms
Convert to operational protocols
Example: Dual-decomposition
Centralized Problem
Introduce Dual
Coefficients
Decompose
These sets cannot be determined
other than at run time!

Automated Decomposition via
Disciplined

Instantiation
Challenges:
How to guarantee one-to-one mapping between instantiated and virtual network elements?
Solution:
Displined Instantiation
Peer random sampling
Hash checking
non-peer
in number of members
conflict instances
Centralized
Control Problem
Centralized
Instance
Distributed
Instances
Apply at
Run-time
Toy Example
Major (and growing) gap
between

Theoretical developments

Practice of wireless networking

Separate communities

Toward an Optimization-Based Wireless Network Operating System
WNOS: A Wireless Network Operating System
Quick Look
Challenges
Network
Abstraction
Time-varying network topology
Abstract before network deployment
Running time information not available
Time-varying neighbor sets,
active/inactive links, interference graphs...
Node

Source of
a session
Destination
of a session
Transmitter
of links
Receiver
of links
Neighbor of
other nodes
Relay of
a session
Cluster member
Cluster head
Consumer of
power, spectrum,
time...
Relay of links
Heterogeneous network behaviors
The same physical device can be different network elements
Network element representation based on directed multi-graph
Bridge network control interface and mathematical models
Abstraction Approach
Three-fold abstraction
Directed multi-graph
Example graphs
Network Control
Problem Definition
Network Abstraction as Directed Multigraph
Network control APIs:
Read, Set, Compose
Scenario 1: Decompose Towards
Social Network Control Objectives
Control Objective:
maximize the social utility of the wireless network by jointly controlling the transmission strategies of different network elements at the link and physical layers
Example Problem: LTE-OPT
LTE-OPT
Optimize the spectral efficiency for LTE networks in the presence of Wi-Fi interference
Each Pico-cell m maximizes its own utility
(a function of channel access time),
with given channel prices
The MBS adjusts the channel prices to minimize the Lagrangian
dual based on observations of the
Pico-cells' channel access strategies
Scenario 2: Decomposition as
Noncooperative Games
Networks are operated by different service providers (SPs) with unaligned network control objectives

SPs are rational and selfish, and their objective is to maximize the capacity of their own networks

Multi-objective optimization problem
Scenario 3: Decomposition for Uncoordinated Networking
Example networks: D2D and V2V networks
Involve all protocol layers of the wireless protocol stack because of the possible multi-hop nature of the wireless networks
Effectiveness
Flexibility
Control programs can be changed
by changing a few lines of code only!

expr = mkexpr('sum(log(wos_x))', 'wos_x')

expr = mkexpr('sum(wos_x)', 'wos_x')
Inflexible
Architectures
Traditional Networking
Architecture
That can't be implemented
in spite of math sophistication...
In Spite of Math Sophistication....
Existing wireless networks are
inherently hardware-based

Innovation lagging behind!

SDRs
are like a CPU without computer architecture, OS, GUIs
What are the right architectures, abstractions?

Based on layering – key to Internet success
Not a good abstraction for wireless
Vertical and Horizontal Coupling of Network Functionalities
Wireless network control is complex
Cross-Layer Control Schemes


Write very complex hardware-dependent cross-layer network control programs

Global network behavior is result of interactions of all different control programs

How About Software Defined Radios?
Traditional SDN
Control
Plane
Considerable utility gain achieved by WNOS

Distributed control with global control objective defined on centralized abstraction

A few lines of code only – distributed and cross-layer optimization programs are automatically generated!
Flexibility &
Scalability
Power Minimization
Rate Maximization
WNOS for Programmable, Optimized Swarm Networking

Distributed Protocols generated by PPS
Network Abstraction
Experimental Testbed
User-defined Centralized
Control Problem
Distributed
Control Problems
Automated
Decomposition
Control Objective
Decomposition
Results
Apply at network run time
J
I
J
i
Control Objective: Find rate at
transport layer (T)
and transmit power at
physical layer (P)
to maximize sum-log-rate
Wireless Cellular Network Operating System (WiCOS)
User's Control Objective
Generate a Network Instance
Decompose
Apply at Run Time
Transmission finished
Increase Rate
Transmission finished
Maintain Rate
WNOS for Secure Wireless Networking
WAVESEC PPS
Built upon RcUBe
May 4, 2018
Catania, Italy

Objective
Bridge the gap between SDN and distributed network optimization/control

Provide
network abstractions
hiding lower-layer protocol details

Hide the details of the
distributed implementation
of the network control operations

Study the principles of a
Wireless Network Operating System (WNOS)

http://www.ece.neu.edu/wineslab/WNOS.php
WiCOS
No Automated Decomposition Tool Available
WiNAR: Wireless Network Abstraction Framework
All Network Designer
Needs to Do
Node A
Neighbors
Node A
Interferers
Example Network Control Problem
A Few Lines of Code Only!
Joint rate and power control
With Zhangyu Guan, Lorenzo Bertizzolo, Emrecan Demirors,
Salvatore D'Oro, Leonardo Bonati

RAN Slicing
WiCOS Objectives
Network Operator
Radio Access Network (RAN) Slicing
Distributed
RAN Slicing
Provide multiple Mobile Virtual Network Operators (MVNOs), i.e.,
the tenants
, with their own virtual RAN , i.e.,
the RAN slice
The RAN consists of multiple
heterogeneous
Access Points (APs)
What is a
good
slicing strategy?
1) Distributed
2) (Near-)Optimal
3) Privacy-Preserving
4) Low-complexity
5) Adaptive
WiCOS Prototyping
Multi-Tenancy Support
Achieved
Results
Ongoing Work
SwarmControl
Map QoS metrics to the optimal operating point in an automated manner, via
Network Management Interface (NMI)
, and configure the parameters of
Programmable Protocol Stack (PPS)
in real time


+
+
Each tenant
individually
selects its own slicing strategy
Each AP has a different
price
Cost function for
tenant
m
to add AP
r

Congestion on AP
r

Monetary cost to
include AP
r

Tenants' users differs in
Number
Position
QoS requirements
Tenant-specific variables and constraints
Tenant-specific control operatives
Optimization Framework
Generation of optimization control problem
Automatic cross-layer decomposition
Final Deployment
Programmable protocol stack solver
Infrastructure enforcement
Interference management and noise cancellation
Orchestration and coexistence of heterogeneous wireless technologies
Multi-Cell Coordination
Highly efficient tenant handover
Tenant-oriented mobility management
Automated Management of Heterogeneous Transmission Techniques
Coordinated multipoint
Cooperative beamforming
Distributed multi-user MIMO
Tenants' strategies are
coupled

High congestion

when everyone includes the same AP in their slice

Slicing strategy
of tenant
m
Slicing strategy
of all other tenants
COUPLING
Our Approach
We use
congestion games
Each tenant
minimizes
the deployment cost:
Minimize
the
congestion
Minimize
the
monetary cost
The game is modeled as a
directed graph
with
parallel links

The game admits a
unique
Nash Equilibrium (NE)

The Price of Anarchy (i.e., a measure of near-optimality) is
bounded by 3/2


The NE can be computed by a
fully distributed
algorithm
The overall slicing cost is no more than 1.5x the cost of an optimal solution
Definition of a distributed near-optimal slicing platform

Include the slicing platform in WNOS to achieve centralized network control through multi-tenant distributed slicing

Development of a working prototype on OpenAirInterface
What we (try to) do in
today's networks
Ongoing Work
http://www.ece.neu.edu/wineslab/WNOS.php
Thanks!
A new approach to control
5G and IoT systems
Zhangyu Guan, Lorenzo Bertizzolo,
Emrecan Demirors, Salvatore D'Oro,
Leonardo Bonati, Tommaso Melodia
Demo of WNOS
IEEE Infocom 2018
Honolulu, HI, April 2018
Full transcript