NFV Architecture Challenges for the 5G Evolution -...

IBM ©2014 IBM Corporation

NFV Architecture Challenges for the 5G Evolution

Dilip Krishnaswamy, IBM Researchhttp://researcher.ibm.com/researcher/view.php?person=in-dilikris

[email protected]

Dell NFV Summit, Santa Clara, CA, USA Nov 17, 2015

http://researcher.ibm.com/researcher/view.php?person=in-dilikris

mailto:[email protected]


The NFV Transition to Software Appliances


NFV

Hardware network appliances are expected to morph into software appliances in data centers

Data centers will be deployed to deliver virtualized network functions

Network functions can be processed in VMs in such data centers

Migrating network functions to software executing on data centers can help reduce capex and opex for network operators and service providers

Typical data center research has focused on compute, storage and energy constraints

NFV Data center performance research needs to explore network constraints as well Source: http://portal.etsi.org/NFV/NFV_White_Paper.pdf

http://portal.etsi.org/NFV/NFV_White_Paper.pdf


What does 5G Seek?

100 Mbps average bandwidth, >10Gbps peak bandwidth

Carry 10,000x more traffic

Support 100x more devices

Low cost M2M

Ultra low latency (1ms) for critical / tactile m2m

Service Agility

Reduce capex/opex for operators

Green networks, devices with long battery life

Increased Reliability and Security

Improve QoE for users

4


Distributed Functions Virtualization (DFV)

5

Cloud Data Center

(Network / Service / App Functions)

In-Network Mini Data Center

(Network / Service / App

Functions)

In-Network Mini Data Center

(Network / Service / App

Functions)

With the availability of compute & storage in-networks, explore theplacement of network / service / application functions across DCs

How can we best utilize the dynamic availability of such distributed compute/network/storage/energy resources?

How can we best place VNFs in hierarchical data centers taking care of latency constraints associated with VNFs, user mobility, energy cost of utilization, resource availability


Example cellular network function partitioning


Hierarchical Partitioning / Function Collapsing

Metric Fully Hierarchical Partially Collapsed Fully Collapsed

Round trip time 179 ms (median) 64 ms (median) 22 ms (median)

Connection setup 3.7 sec (median) 1.3 sec (median) 0.7 sec (median)

TCP bandwidth 3.19 Mbps 3.45 Mbps 3.72 Mbps

Playout Stalls 12 2 0


Distributed Function Virtualization (DFV)

NaaS moves networks into cloud

DFV moves data centers hierarchically into networks

DFV across Hierarchical NFV POP (Point Of Presence) Data Centers

(DCs) using Interconnected Distributed VMs

Placement of VMs in the hierarchy to meet performance requirements

Flexibility of VM placement in the NFV POP data center hierarch

Collapsed Function Virtualization Systems

Hierarchical DC Resource Mgmt, Policy Mgmt, & Analytics

VM Edge Applications Enabling on NFV Data Centers

–Can Enable Higher QoS/QoE Apps: VoiceOverLTE & VideoOverLTE

–Content delivery, Microcloud apps, Gaming, Augmented Reality


Level 3 Data Center ( L3DC)

Level 2 Data Center (L2DC)

SFVM NFVM

Level 2 Data Center (L2DC)SFVMSFVM NFVMNFVM AFVMAFVM

Level 1 DC (L1DC)

NFVMAFVM

SFVM

Level 1 DC (L1DC)

NFVMAFVM

SFM

AFVM

Level 1 DC (L1DC)

NFVMAFVM

SFVMLevel 1 DC (L1DC)

NFVMAFVM

SFVM

HFM

HFM HFM

HFM HFM HFM HFM

User

Device

Network Access Device Network Access Device Network Access Device Network Access Device

User

Device

User

Device

User

Device

User

Device

User

Device

User

Device

User

Device

General Distributed Function Virtualization


Level 1 DC (L1DC)VMs: ENodeB-NF, RNC-NF,

CDN-SF, AFs


CDN-SF, AFs


CDN-SF, AFs

Level 2 DC (L2DC)

VMs: SGSN-NF, S-GW-NF, MME-NF,

CDN-SF, AFs

Level 3 DC (L3DC)VMs: GGSN-NF, P-GW-NF, LB-NF, FW-NF, NAT-NF, MTC-IWF-NF,

M2M-SF, CDN-SF, IT-SF, AFs

Level 2 DC (L2DC)VMs: SGSN-NF, S-GW-NF, MME-NF,

CDN-SF, AFs


CDN-SF, AFs

HFM

HFM HFM

HFM HFM HFM HFM

User

Device

Network Access Device Network Access Device Network Access Device Network Access Device

User

Device

User

Device

User

Device

User

Device

User

Device

User

Device

User

Device

3G/4G Distributed Function Virtualization


Hierarchical Function Manager (HFM)

HDCPM

HRM

HAE

HIOM

HFM

HFM : Hierarchical Function Manager

HIOM : Hierarchical I/O Manager

HAE : Hierarchical Analytics Engine

HDCPM : Hierarchical Data Center Policy Manager

HFM : Hierarchical Resource Manager


Hierarchical Latency & Policy aware Partitioning - InterDC

Example Depiction – Serving users in Sunnyvale and Cupertino CA

ILOG CPLEX-based Partitioning

Cupertino

Small DC

San Francisco DC

Arizona DC (Solar Powered)

Sunnyvale

Small DC

Low Energy cost, Higher Latency

High Energy cost, Medium Latency

Lower Latency, Lower Capacity

High Energy Cost

Dynamically divert new VNF resource requirements away from a DC if heavily loaded

Map users to VNFs on DCs based on latency, availability, DC load, energy, mobility

Dynamically direct new user flows to utilize VNFs at the most appropriate DCs

IEEE NFV-SDN 2015


NFV Data Center Resource Management and Orchestration

NFV data center research needs to explore compute, network, energy, storage, and latency constraints, and in a distributed manner as well

Both IntraDC and InterDC resource management and orchestration need to be addressed


IntraDC Resource Management

14

IEEE SETCAC 2015


Hierarchical Data Center Resource Management

Intra-Data-Center and Inter-Data Center optimizations

Dynamic monitoring

Dynamic resource management

Placement of VMs within and across data centers

Consolidation of soft and hard constraints based on latency,

energy, compute, networking, storage, utility constraints

–Understand feasibility regions based on constraints

–Explore policy admission or relaxation of policies to ensure

an adequate region for operating conditions

15


Co-existence with existing appliances


IBM OSS for Cloud Based Networking High Level Architecture

IBM Service Design & Creation IBM Dynamic Lifecycle Orchestration IBM Operational Analytics

Design ToolsIBM

Urbancode/BPM

AssuranceIBM Netcool

AnalyticsIBM Now Factory

SecurityIBM QRadar

NFV Orchestrator

Virtual Infrastructure

Manager

NFVI

Resources

Real

Time

Policy

vnf1vnf2

vnf3

Legacy NetworksLegacy Networks

EMSLegacy

OSS

• Service Specs

• Assurance

KPIs

• Policies

• VNF

Descriptors

Service/VNFCatalogs

NFVI - NFV Infrastructure

Compute ResourcesCompute Resources

Virtualization Layer: KVM; VMWare ; VMControl

Storage ResourcesStorage ResourcesNetwork Resources

(Switches and Routers)

Network Resources

(Switches and Routers)

Virtual

Compute

Virtual

Storage

Virtual

NetworkSDN

Controller

SDN

Controller

Legacy OSS/BSS

IBM Rational

Rapidly onboard VNFs and design service Orchestrate service instances and auto configure operational runtime

Real time self healing,

scaling & movement


Hierarchical Function Splitting

18


ADREMO - Analytics Driven REsource Mgmt and Orchestration

19

• Static/Dynamic/Autonomic Constraint/Policy/Resource Mgmt

• VM / Container management / orchestration

• InterDC VM Placement Mgmt

• ADREMO-L (IntraDC) , ADREMO-W (InterDC)

IntraDC

Compute

SS

ADREMO-L

Storage

SS

Network

SS

Pub/Sub Bus

Monitor

Extract

Learn

Predict

Optimize

Orchestrate

DC3DC2DC1

Pub/Sub Bus

InterDCADREMO-W


Local microPCRF / microPCEF

App/VM/container(s)

Local Access

Node

Remote

PCRF

Remote

PCEF

UE

Local

Gateway Node

Traffic to/from UE

Distributed Billing with Route Optimization


Hierarchical vCDNs


Viral Video Analytics

IEEE CCNC 2015.


Distributed Analytics

IEEE CCNC 2015.


Average power savings (dynamic vs static)

Wdiff = Wtotal,static - Wtotal,dynamic* (1 + γ) = Widle,static - Widle,dynamic

Widle,static - Widle,dynamic = (120 – 10) = 110

Average power savings = Wdiff * Pidle = (120 – 10) * 100 = 11000 Watts.

Percentage Power savings = (11000/40500)*100 ≈ 27.16%

IEEE CCNC 2015.


Switch Handoff across Controllers

25

Select best Controller based on link conditions vs controller load

IEEE ANTS 2015


Cyber Threat Resilience for Virtualized Infrastructure

Enable dynamic adaptable defense systems in both control and

user/data planes

Continuously monitor systems and reprogram systems on-the-fly

Disable faulty VMs/containers, and redirect to new fixed

VMs/containers

DDoS mitigation

Quickly throttle or disable resources for flows/services/applications

Monitor data access to maintain data Integrity and confidentiality

Authentication of Virtual Functions prior to dynamic instantiation

26


Cyber Threat Resilience for Virtualized Infrastructure

Authentication of orchestrators for Virtual Functions

Prevent Unauthorized Virtual Functions

Monitor suspected intrusions, Configure Firewalls

Dynamically Update IP blacklists

Authorized packet access / modification by VNFs

Authorized Virtual Service Functions to provide service

Dynamic warning systems for Cyber Threat Resilience

Distributed Consensus mechanisms for Cyber Threat Resilience

27


Virtual Apps (vApps) - WWW 2015 conference

NFV-enabled platform to host app-servers inside

operators network

QoE


Signaling Proxy (vProxy) - WWW 2015 conference

NFV-enabled platform to host operator services to

manage scale of mobile phones and mobile apps

and m2m devices


Joint NFV / SDN Evolution

Explore new innovative paradigrms for SDN/NFV

– The interplay between SDN and NFV can be interesting

– Distributed & hierarchical architectures for both SDN & NFV can be interesting

and relevant

For SDNs, can think of novel partitioning between capabilities in hardware

appliances and virtualized software components

– Not necessary to think about an explicit partitioning between control and user

planes

– For control – explore hierarchical control

– Exploring SDN mechanisms to better enable NFV

For NFV

– Explore interesting virtual network functions that can be hosted in data centers

– Leverage distributed compute/storage to deploy new virtual network or service

or application functions

– Leverage SDN to help with hierarchical and distributed orchestration of VNFs

– Explore new architectures and solutions based on NFV and SDN

• Explore distributed computing, networking, energy and storage to provide

improved services

30


WiFi APNetwork Access

Device

Generalized DFV Network, Service, and Application Function VM Partitioning

User 1 Application

L3DC

L2DC

L1DCEnodeB

VMS-GW

VMP-GW

VM

M2M

Serv

ice

Laye

rS

F V

M

CDC

Internet AccessProvider VM

User 3 Application

Inte

rO

pera

tor

Tunnel

SF

VMP-GW

VMS-GW

VMEnodeB

VM

App

Pro

xy

Serv

er

SF

VM

Ap

p2

Serv

er

VM

L2DC2

Network Access Device

User 2 Application

P-GWVM

S-GWVM

App

Serv

er

VM


DFV Summary

Utilize Distributed Function Virtualization (DFV)

Techniques in 5G NFV Architecture to complement 5G

physical layer transformations –Optimized collapsed connectivity can provide lower

latencies

– Programmable and optimized Control and User/Data

Planes can improve great flexibility

–Concurrent connectivity across multiple paths for the same

flow or for different flows can help improve performance

and optimize QoS/QoE across flows and across users and

networks

– Software appliances can be moved and placed as needed

to deliver the best performance in the network

32


Overall Summary

5G DFV Architecture can enable distributing and connecting

functions in a hierarchical and distributed manner at the edge,

in the network, and in the cloud providing dynamic agile

deployment and scaling options

Hierarchical and Distributed Function Partitioning, Collapsing,

Replication, and Splitting can be key areas of optimization for

enabling DFV in 5G NFV systems

Need to provide resilience to potential security threats

Functions can be connected together and placed in interesting

ways to create new architecture and new solutions for people

5G is new area we can collectively dream and define what

we want it to be, for 5G systems to do the best that they can, to

serve people better

33


Related Articles

D. Krishnaswamy, R. Kothari, V. Gabale, "Latency and Policy aware Hierarchical

Partitioning for NFV Systems", First IEEE NFV-SDN conference, 2015.

IBM OSS for Cloud-based Networking Whitepaper http://www-01.ibm.com/common/ssi/cgi-

bin/ssialias?subtype=WH&infotype=SA&htmlfid=MSL03006USEN&attachment=MSL03006USEN.PDF, 2015.

D. Krishnaswamy, Forbes blog on DFV, http://forbesindia.com/blog/business-strategy/bridging-the-

digital-divide-with-national-knowledge-centres/, 2015

R. Krishnan, T. Hinrichs, D. Krishnaswamy, R. Krishnaswamy, “Policy-based

Monitoring and Energy Management for NFV Data Centers”, IEEE SETCAC 2015.

Y. Malreddy, D. Krishnaswamy, B. S. Manoj, "Cross-Layer Switch Handover in

Software Defined Wireless Networks," IEEE ANTS 2015.

D. Krishnaswamy, R. Krishnan, D. Lopez, P. Willis, A. Qamar , "An Open NFV and

Cloud Architectural Framework for Managing Application Virality Behavior", CCNC’15

D. Krishnaswamy, “Cyber Threat Resilience in Virtualized Infrastructure”, Invited talk,

International workshop on CyberThreat Resilience, 2015.

V. Gabale, D. Krishnaswamy, "MobInsight: On Improving the Performance of Mobile

Apps in Cellular Networks", 24th International World Wide Web conference 2015.

34

http://www-01.ibm.com/common/ssi/cgi-bin/ssialias?subtype=WH&infotype=SA&htmlfid=MSL03006USEN&attachment=MSL03006USEN.PDF

http://forbesindia.com/blog/business-strategy/bridging-the-digital-divide-with-national-knowledge-centres/


Thank you

35

NFV Architecture Challenges for the 5G Evolution -...

Documents

Transcript of NFV Architecture Challenges for the 5G Evolution -...