Cross-Layer Optimization (CLO) Problem Statement Young Lee, Susan Hares, Greg Bernstein.

Cross-Layer Optimization (CLO)

Problem Statement

Young Lee, Susan Hares, Greg Bernstein

Page 2

Cross-Layers Application Layer (Host Layer) Distributed Resources: servers, content,

data sets, computing power, cache/mirror Uses Network Resources Different QoS requirements for each

application

Transport to Network Layer Bandwidth, Connections, Links, Connection Processing (Creation, Deletion,

Management) Admission Control, Resource Reservation Applications uses resources in IP, MPLS,

and/or Optical Transport Networks, Layer 2

Transport Layer

Network Layer

Application Layer

Page 3

Applications need Assured performance levels associated with

the application. Optimized resource utilization via cross-layer

interaction between application and network

layers, A cross-layer management paradigm that is

dynamic, interactive, elastic, adaptive, and

flexible across application-network layers.› Elastic – stretches the network during growth

› Adaptive – changes with network

Page 4

Current and Emerging Application Resources

Live Data Sources› Video or audio from live sporting or entertainment events, data feeds

from radio telescopes, remote medical surgery

Processing Resources› Raw computational capability for cloud computing, transactional

capabilities for e-commerce, processing for streaming media,

transcoding capabilities for video and audio, etc...

Storage Resources› Disk farms, tape libraries, etc...

Content/Data Sets› Video, audio, commercial, scientific, etc...

Page 5

Application Service ProfilesCharacteristics & QoS Requirement of application

service from a network perspective: Location profile: locations of both the clients and the sources QoS profile: (i) Delay Tolerance Bound; (ii) Jitter Tolerance Bound; (iii) Packet

Delivery Ratio Tolerance; (iv) Network Availability, etc. Connectivity profile: (i) P-P; (ii) P-MP; (iii) MP-MP; (iv) Any Cast Directionality profile: (i) uni-directional; (ii) bi-directional Bandwidth profile: Maximum, average, and minimum bandwidth

requirements for the connectivity, maximum burst rate, maximum burst duration, etc.

Duration of service profile: service time of the application Network media profile: (i) optical only; (ii) no microwave, etc. Restoration profile: (i) Reroute required; (ii) do not re-route, etc. Security profile: (i) dedicated end-to-end VPN-like resource allocation; (ii)

dedicated physical resource allocation

Communicate these application profiles to network via a common mechanism between application and network layer

Page 6

File/Content Distribution SystemsDownload of images/audio/video/software via the network;

Common optimization problems in this system includes:

Cache and Mirror placement problem› Coordination of the application and network (transport and IP) topological

information is key to optimization

Efficient transfer of content to servers› Coordinated point-to-multipoint concurrent path optimized with network

loading condition in both application and network is key to optimization

Client to server assignment problem › Current server load and network latency between client and server are

key QoS

Page 7

Content Distribution Network (CDN)

Surrogate

Surrogate

Surrogate

CE

CE

CE

User User

User User

User User

CE CE

CE CE

CE CE

PEPE PE

PE

PE

PE

DNS server(ANC)

TNC Transport NetworkService request

CDN NetworkDNS server performs the ANC function. • It receives users’ requests

• Based on surrogate’s availability, transport network resource and user’s position, it chooses the best surrogate to serve the user.

• It sets up a PE-to-PE connection between the surrogate and the user through communication with the TNC in the transport network.

• The users who share the same surrogate and the access PE can share the PE-to-PE connection.

• B/W modification capability is also needed

Page 8

Streaming Content Distribution Systems

The streaming case increases the need for coordinated multi-

layer monitoring and configuration primarily due to

more stringent QoS constraints on bandwidth and

jitter.

Optimization problems for a live streaming service include:› Server selection and placement problems (application based multi-cast)

› Leaf attachment problem and tree construction (network-based multi-

cast)

Additional optimization decisions required with on-demand

streaming› Client stream sharing

› Batch or multicast server selection problem

Page 9

Video On Demand

ContentSource

SuperHead

End (SHE)Location 1

SuperHead

End (SHE)Location 2

ContentDistribution

Local AreaHead End

1

End User

PEPE

PEPE

Local AreaHead End

2

End User

Local AreaHead End

3

End User

PE

CE

CE

CE

CE

CE

VoD ServiceController

(ANC)

ApplicationCloud

• VoD Service Controller (VSC) performs the ANC Function• Local Area HE sends a query the VSC for video download• VSC makes a decision which SHE should send the video to the Local Area HE based on:

• Transport Network Topology (that TNC provides) and Server (SHE) status and Movie Availability

• QoS: Delay and Jitter sensitive• Connection: One-way

1. service request

Transport Network

TNC

2. VNT3. Reserve

3. Download request

ANC – Application Network ControllerTNC- Transport Network Controller

Page 10

Conferencing and GamingThese applications increase the complexity of the overall

application connectivity and the need for cross-layer

coordination of monitoring, configuration.

Bi-directional connections and asymmetric bandwidth

between the server and the user location Multipoint-to-multipoint connectivity with hard QoS

constraint on latency and bandwidth › Data path formation and reformation for MP-MP can be very inefficient

without considering the underlying network resources

› Network path computation and path reservation may be required to

ensure the end-users service objective.

Gaming adds additional scalability on QoS

requirement and the connectivity.

Page 11

Video Conferencing Architecture

Meeting RoomSource

CE PE

Meeting RoomDestination 1

Meeting RoomDestination 2

Meeting RoomDestination 3

CE

CE

CEPE

PE

PE10 Mbps

30 Mbps

Meeting RoomDestination 4

PE

CE

Video ConferencingController

(ANC)

• Video Conferencing Controller (VCC) performs the ANC function. • It receives client requests • It sets up the P-MP connections via the TNC in the transport network• QoS: Delay and Jitter sensitive• Connection: Asymmetric P-MP connections (uplink: 10 Mbps; downlink: (N-1)*10 Mbps where N is the number of participants)

• Desk Top conferencing is envisioned to connect 100’s of clients which will require higher B/W.

Service request

TNC

ANC – Application Network ControllerTNC- Transport Network Controller

Page 12

Video Gaming Service Architecture

GameServer

GameServer Game

ServerCE

CE

CE

Gamer Gamer

Gamer Gamer

Gamer Gamer

CE CE

CE CE

CE CE

PEPE

PE

PE

PE

PE

Video GameController (ANC) TNC Transport Network

Service request

• Video Game Controller (VGC) performs the ANC function. • It receives client requests • It sets up the P-P connections via the TNC in the transport network• Server location and B/W availability from Server to Client are key info for decision• B/W modification capability is critical• QoS: Delay and Jitter sensitive• Connection: Asymmetric P-P/MP-MP connections (uplink: 100 Mbps; downlink: 10 Mbps)

• Up to 100,000 clients connections (E.g., Worldwide war games)

ANC – Application Network ControllerTNC- Transport Network Controller

Page 13

Grid Computing/Remote Medical Grid computing supports extremely large transfers of files and

streaming data. The volume of the traffic makes it critical to

synchronize changes to application and network. Key issues with Grid Computing include:

› Instantiation of the connectivity with high data rates (100’s Gbps) and/or

data set size (1000’s)

› Controlling very high speed network

Remote medical application adds more complexity over Grid

Computing in that it requires fast setup of 100Gbps level of

connectivity with higher security and more stringent jitter

requirements including MP-MP connectivity. Lambda-level optical transport with control plane would

be required with a sophisticated path computation

algorithm across layer. (Optical Bypass application)

Page 14

Problem StatementThe lack of common coordination mechanism between

the application and each of the layers in the

“network” does not allow coordinated cross-layer

optimization:

Coordinated “query” of application and network

requirements to determine available computing and network

resources (including resource availability and demand as a

function of time); Coordinated provisioning processes (resource reservation) of

both application and network layers based resource

availability on both layers (statistical and/or real time); Coordinated cross-layer monitoring; and Quick re-optimization based on policy of the

application/network upon churn

Page 15

Existing Solutions Only Provide Partial Solutions IETF Management solutions: SNMP, Netconf/Yang,

› Do not provide the necessary context to view across multiple layers, multiple

devices/technologies.

› Lack of a context that allows synchronization of actions for read-view, write-

view, notify-view and actions.

MPLS OAM› MPLS OAM is limited to MPLS device. The current scope of MPLS OAM does

not support non-MPLS devices for its configuration and provisioning functions.

› Lacks topology sharing and configuration interface.

ITU-T Y.2011/2012› Defines application network interface (ANI); however, it does not address any

details on cross-layer synchronization of information, configuration and

provisioning

IETF ALTO WG › Current scope does not address the multi-layer synchronization problems

› Does not provide the mechanism to configure/re-optimize and provision

across layers.

Page 16

SNMP Example

CECE

PEPE NetworkIngress

NetworkEgress

Data path

BridgeMIB L2 Core

Transport Network

L2

ApplicationSource

ApplicationSource

End PointConsuming ResourceEnd Point

Consuming Resource

ApplicationSource

L2VPLSMIB- PE

L2VPLSMIB-

BridgeMIB

BridgeMIB

L2VPLSMIB-

TransportMIB

TransportMIB

TransportMIB

Roll-forward, Roll-back

For massive data or time critical

Implies synchronization or

Atomic handling

Potential Solutions

Page 18

Cross-Layer Optimization Functions (Enablers) CLO functions: To exchange network capabilities

(Network Capability Exchanges)

To initiate service instantiation of application to network with profile exchange (Provisioning)

To exchange topology and/or traffic-engineering related information between the layers (Coordination of cross-layer D/B’s for joint optimization)

To exchange application/network congestion/failure information (Coordination of configuration changes)

CLO functions

Page 19

Information sharing and CLO entity ownership

Information to be shared across layer depends on CLO ownership› CLO by application provider Sharing of some type of topology information

› CLO by network provider sharing of some type of application information

› Other possibilities include neutral 3rd party “broker”

CLO functionsCLO functions

CLO Entity

CLO Entity

3rd PartyBroker

Page 20

Generalized Data Path Connectivity (Single Domain)

CECE

PEPE NetworkIngress

NetworkEgress

Data path

Core Transport Network(L1 or L2 or L3/MPLS-TE)

CLO Functions

ApplicationSource

ApplicationSource

End PointConsuming ResourceEnd Point

Consuming Resource

Page 21

Multi-Layer, Multi-Device, Multi-Domain Scope of CLO

Carrier 1IP L3 Based

Carrier 2L2/SPBB Based

Carrier 3 OTN/WDM

Application

WDM (ROADM) OTN/WDM

Multi-Domain Adaptation and Translation

Cross Layer Communication

Page 22

Summary – Proposal to IETFCreate an IETF WG to: Define Cross-layer optimization functions and

architecture Develop protocols that would enable:

› Information Exchange between Application and Network Layers

» Topological and TE (Virtual/Abstract) view of each layers to be

exchanged per query

» Monitoring data exchange

› Allow the Application Layer to request for a path estimation (path

feasibility) to Network Layer

› Allow the Application Layer to initiate a provisioning connection/flow to

network layer

› Multi-layer, multi-technology adaption and translation

› Potential multi-domain

Page 23

Thank You