HOW A PARADIGM SHIFT IN NETWORK ARCHITECTURE MAY IMPACT ENERGY EFFICIENCY (SERVICE VIRTUALIZATION) Spyros Denazis (sdena@upatras.gr), University of Patras, GREECE

Ημερίδα ENDECON (Θαλής)

A broader holistic vision

SDN Device

Programmability

Service Function Chaining

Service Lifecycle Deploy (NFV) Integrate (SDN) Connect (SFC) Control (SDN) Destroy (NFV) Network

Functions Virtualization

These concepts have a past, therefore, no need to re-invent the wheel as there have been lessons already learned

Service Instantiation

Datapath Element DE

SP

DE DE

SP

SP

SP

SP

Service Part

Control Element Management/Orchestrator

Deploy Instantiate Control Interconnect

Control Manage

App App App App Publish

Service Instantiation

Datapath Element DE

SP

DE DE

SP

SP

SP

SP

Service Part

Control Element Management/Orchestrator

Deploy Instantiate Control Interconnect

Control Manage

App App App App Publish There is a need for a common framework that will enable the control/management/orchestration of the network substrate to be treated in uniform fashion and simplify their corresponding operations

Network Element (NE)

Control Plane

ForCES Framework

Control Element (CE) Control

Element (CE) Control Element (CE)

Forwarding Plane

Forwarding Element (FE) Forwarding

Element (FE) Forwarding Element (FE)

ForCES protocol

Network Element (NE) Packet Processing Entity Constitutes of CEs & FEs

Multiple CEs to FEs for HA CEs/FEs Physical or Virtual

NE components distributed Local (within one box)

Geographical distributed (LAN/WAN/Internet)

No discrimination between physical or virtual

Forwarding Element (FE)

ForCES Abstraction model

Control Element (CE)

LFB1 LFB2

LFB3 LFB4

LFB3

LFB4

Model FEs using Logical Functional Blocks. Fine grained operations Hardware/Software Physical/Virtual

FE - graph of LFBs Graph can be dynamic if supported

by implementation Model Object-oriented approach

Classes Instances Etc

P: Packet M: Metadata

P P+M

ForCES

An Initial Proof of Concept based on LTE

Questions / Motivation

Can we apply SDN concepts in the EPC Core? Separate Control/Data Plane

Proof of concept: Use ForCES on the PGW

Backward compatibility with 3GPP

Motivation: Scale-in & Scale-out PGW

Ability to add functionality as LFBs wherever possible. Chain LFBs to perform new services.

E.g. Firewall/DPI LFBs to the PGW

Group and co-locate Control Plane functionality

Simplified PGW viewpoint

PGW

SGW

Policy Cloud

PGW - D

PGW - C

Internet

GTP-C

GTP-U

ForCES

Separate PGW control/datapath Current focus:

Create tunneling endpoints for connection with Internet Both for SGW & PGW

Collect usage statistics

PGW Prototype Implementation

PGW - D

PGW - C

ForCES

Port LFB GTP-U LFB Port LFB

Use case 1- Incoming Packets from SGW

PGW - D

PGW - C

ForCES

Port LFB

GTP-U Encap

Port LFB GTP-U Decap

1 P’

2 P’

3 P

4 P

P’: GTP-U encapsulated packet P : Initial Packet

PGW - D

PGW - C

ForCES

Port LFB

GTP-U Encap

Port LFB GTP-U Decap

4 P’

3 P’

2 P

1 P

P’: GTP-U encapsulated packet P : Initial Packet

Use case 2 - Incoming Packets from Internet

Proof-of-concept prototype testbed

GTPvU In/Out (SGW)

Host 2 (Internet)

Host 1 (UE)

GTPvU In/Out (PGW)

PGW - C

GTPv1-U

SGW - C

ForCES

IP IP

App creating tunnels/collect stats

Proof-of-concept Sequence

1. VM start-up 2. Initialize PGW/SGW 3. Create Tunnel Endpoints for Hosts 4. See traffic flow 5. Collect statistics Per UE Per UE/Flow

Future Research Directions

SDN has grown beyond simple programmability (control) NFV + SFC

Service granular decomposition and deployment are enabled and they provide key advantages Generalize it also to Control & Management Planes

Orchestration & Resource Management will be of paramount importance They emerge as the next step forward

Is there a need for a common framework & abstraction model (holistic view) ?

Or, a collection will guarantee convergence?

Potential Impact on Energy

The ability to scale-in, scale-out on demand network functionality and localize it allows for energy optimization in various ways Use existing or future data centers policies/approaches Regulate energy consumption according to demand Apply more efficient groupings of the functional

components of an architecture

The potential of these new network architectures may have on energy efficiency has not been investigated thoroughly

http://nam.ece.upatras.gr/

ΤΗΑΝΚ ΥOU!

Backup Slide – PGW model details

GTP tunnel characteristics UE identification (IP address in our case)

Local Tunnel Endpoint IP address

Remote Tunnel Endpoint IP address

Local Tunnel Endpoint ID

Remote Tunnel Endpoint ID

GTP Statistics Bytes Sent/Received

Packets Sent/Received

GTP Deep Packet Inspection Statistics GTP Statistics Per Flow

Flow characterized by 5-tuple: Source/Destination IP Address

Source/Destination TCP/UDP port

IP Protocol Type