[PPT] - TIGER T ogether I P , G MPLS and E thernet R econsidered Carrier PowerPoint Presentation

SLIDE 1

TIGER

Together IP, GMPLS and Ethernet Reconsidered

Carrier Ethernet: Identifying and Avoiding Remaining Stumbling Blocks

Euroview 2008 - Wurzburg, July 21-22, 2008

The TIGER Consortium Project Coordinator: Nicolas Le Sauze (Alcatel-Lucent France) Presenter: Dimitri Papadimitriou (Alcatel-Lucent Bell) <dimitri.papadimitriou@alcatel-lucent.be>

SLIDE 2

2 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

Outline Introduction (CELTIC TIGER Project) Problem statement: Motivations and Research challenges Carrier Ethernet technologies - Positioning Carrier Ethernet technologies - Performance benchmarking Carrier Ethernet technologies - Cost benchmarking Deployment scenarios Main research directions

SLIDE 3

3 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

Who we are ?

1 operator 3 equipment vendors 2 SMEs involved in ICT business 2 research institutes 2 universities

Duration: 2 1/2 Year (March’05-June’08) Our objectives ?

Orange – FT group

CELTIC TIGER Project

Propose solutions for a better coordination between IP and Ethernet technologies to address the Metro Ethernet growing market Propose solutions for a better coordination between IP and Ethernet technologies to address the Metro Ethernet growing market Evaluate by experimentation (simulation and emulation) benefits of the proposed solutions with regard to existing technologies on the market Evaluate by experimentation (simulation and emulation) benefits of the proposed solutions with regard to existing technologies on the market

SLIDE 4

4 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

Motivations

1. Metro = strategic link between access & core
Traffic increase (amount & rate): 50-70%
Progressive bottleneck (capacity and service)
New constraints:

Large amount of traffic (variable) Aggregation close to final users Flexible resource allocation

Source: European Commission Sixth Framework Integrated Project BROADWAN

2. Ethernet increasingly attracting service providers' for metro networks
Ethernet HSI together with CAPEX reduction (1GbE, 10GbE and by 2010, 100 GbE)
Multi-service edge routers interconnection (EPL, EVPL, EPLan, EVPLan)
Aggregation technology of choice in metro networks (supposedly low cost)
3. Unsuitable and complex architecture inheritance
Not adapted to existing infrastructure (SDH, ATM/FR, etc.)
Lack of interoperability, complexity, and too expensive (OPEX e.g. static config)

⇒ New network technology & architecture: Carrier Ethernet New service and operational needs: Key differentiators in terms of CAPEX & OPEX

SLIDE 5

5 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

Challenges

Paradox: moving Ethernet "networking" properties (associated to LAN / campus) toward

metro networks - but also core - would definitely transform intrinsic nature of Ethernet ⇒ Improvements required to address carrier class requirements e.g. scalability, traffic engineering, recoverability, and manageability

Provisioning (Forwarding Components) Provisioning (TE data paths, re-routing, etc) Forwarding control

Forwarding Control (MSTP) Forwarding Control (GMPLS) Management Management

Spanning Tree, Learning, Filtering Provisioning (Policy, etc) Provisioning (Forwarding Components)

Two groups of Carrier Ethernet solutions: Ethernet centric (PBB, PBB-TE, etc.) and IP/MPLS centric (PW/MPLS, MPLS-TP, etc.) – Which orientation ? Ethernetization of MPLS -or- MPLSization of Ethernet ? – Does one size fits all ? – Scaling, cost/functionality, cost/gain, & cost/performance ratio ?

SLIDE 6

6 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

Carrier Ethernet Technologies - Positioning

PBB-TE PBB-TE VLAN Bridged (PB - 802.1q/.1ad) VLAN Bridged (PB - 802.1q/.1ad) VLAN Bridged + MiM encaspulation (PB/B -802.1ah/.1ad VLAN Bridged + MiM encaspulation (PB/B -802.1ah/.1ad Ethernet Label Switching Ethernet Label Switching

LAN/Campus LAN/Campus Access Access Metro Network Metro Network

Carrier Ethernet:

1. Adapt forwarding components

(for Ethernet switching)

2. Depart from IEEE 802.1 (xSTP)

control components Multiple Spanning Tree Protocol → Constraint-based routing for TE data path establishment Rapid Spanning Tree Protocol → Fast Re-routing MAC learning (flooding) → discovery (link state routing protocol) Carrier Ethernet:

1. Adapt forwarding components

(for Ethernet switching)

2. Depart from IEEE 802.1 (xSTP)

control components Multiple Spanning Tree Protocol → Constraint-based routing for TE data path establishment Rapid Spanning Tree Protocol → Fast Re-routing MAC learning (flooding) → discovery (link state routing protocol)

xSTP / 802.1 control plane lacks CG features such as TE, recovery, etc.

IP Control Plane

Scalability Carrier grade

Switched Ethernet

Bridged Ethernet (PB/B - 802.1d/.1q) Bridged Ethernet (PB/B - 802.1d/.1q) 5km 50-100km

SLIDE 7

7 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

xSTP -> Lack of TE and fast-re-routing / fast- convergence (O(10-100ms)) Relies on MAC frame lookup, flooding and learning Complex MAC frame processing at edges (PBBN) Forwarding Independent of client MAC address (MAC tunnels) No VID space limitation (as service delimiter) OAM (802.1ag) PBB May require label merging usage (to circumvent per link label space limited to 4K) Single level trunking technology (not a limitation when used within single domain) Link-state routing protocol Traffic engineering (p2p and p2mp) Both end-to-end & local/segment fast re-routing Simple processing at network edges (G)ELS No local/segment fast re-routing (re-merging is impossible) Complex MAC frame processing at edges (PBBN) Requires shared forwarding (multiplexing) to prevent limitation of 4K data paths per destination Intermediate nodes require upgrades Support of multicast comes with additional complexity Same as PBB + Traffic engineering (point-to-point) + Mis-merge traffic is not propagated PBB-TE Scalability and performance because of MAC frame lookup, flooding and learning xSTP -> Lack of TE and fast-re-routing / fast- convergence (O(10-100ms)) Widely available in non-carrier environments xSTP

Con’s Pro’s Techno

Which native carrier Ethernet technology

SLIDE 8

8 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

(G)ELS – Technology

Topology: single domain
Data path: p2p, p2mp
Mode/Fwd: switch/S-VLAN ID (S-VID)
OAM: BFD (native)
Service boundary/Type: none
Properties:

VID translation (swap operation) is standard (802.1Q) EtherType (-> new features can be easily introduced) MAC address independent forwarding

Switching: 4096 paths per link

E-LSE E-LSR E-LSE Source Dest Router S-VID swap S-VID push S-VID pop Ethernet LSP Ethernet MAC frame Ethernet MAC frame Router

Source MAC Dest. MAC Control Control Control

ELS Building Blocks ELS Building Blocks Ethernet 802.1ad (S-VID translation) Ethernet 802.1ad (S-VID translation) RSVP-TE OSPF-TE RSVP-TE OSPF-TE Control component Forwarding component GMPLS BFD

Establish and maintain Ethernet LSP (Label

Switch path) using the VID though E-LSR

VID is translated at each LSR, VID is link local
Control plane and OAM using (IETF tools:

GMPLS protocol suite and BFD)

SLIDE 9

9 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

MTN – Technology

Topology: multi-domain
Data path: p2p, p2mp
Mode/Fwd: switch/label
OAM: Transport-centric OAM (G.8114-like)
Service boundary/Type: PW label/Multi-Services
Properties: MPLS based (but decoupled from IP)
Open points:

Addressing Make MPLS a layer 2 technology is challenging Interworking PW/MPLS

Establish and maintain MTN LSP (Label Switch

path) using the label swapping

PW/MTN Label is swapped at each PE/LSR
Control plane (IETF tools: GMPLS protocol

suite with ASON extensions) N:1 N:1

MTN Network MPLS Network

GMPLS RSVP-TE RSVP-TE/LDP T-LDP

MPLS PW MTN “PW” MPLS tunnel MTN “tunnel” MPLS tunnel

T-LDP T-LDP

SLIDE 10

10 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

MTN ELS TIGER focused on defining and completing two new technologies: ELS (Ethernet Label Switching) and MTN (MPLS for Transport Networks)

MTN Innovation Space Adaptation of GMPLS and ASON control protocols Enhanced provisioning abilities for pseudo-wires including multicast support Transport-centric OAM (G.8114- like) MTN Innovation Space MTN Innovation Space Adaptation of GMPLS and ASON control protocols Enhanced provisioning abilities for pseudo-wires including multicast support Transport-centric OAM (G.8114- like) ELS Innovation Space Multi-class and multi-traffic data path based on Ethernet LSP capabilities Label merging to elevate label value space limitation BFD adapted to Ethernet needs (no dedicated hardware as with IEEE 802.1ag) ELS Innovation Space ELS Innovation Space Multi-class and multi-traffic data path based on Ethernet LSP capabilities Label merging to elevate label value space limitation BFD adapted to Ethernet needs (no dedicated hardware as with IEEE 802.1ag)

TE capabilities (incl. data path provisioning and recovery) Use and take advantage of local labelling Forwarding decision in transit LSRs independent of technology addressing schemes (e.g. MAC DAs)

TIGER Solutions

SLIDE 11

11 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

Benchmarking: Cost, Dimension, & Performance

Objectives: benchmarking of the proposed architectures/technologies using set of scenarios/case studies Main problems to be solved: solution value

Dimensioning studies: evaluate cost/gain ratio to deliver feature (set) Performance studies: evaluate cost/performance ratio to deliver feature (set)

Technical approach:

Dimensioning studies: gain objectives (data & control plane resources) Performance benchmarking studies: performance

bjectives (data & control plane resources)

Benchmarking Performance Cost Functional

SLIDE 12

12 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

OAM performance

5 6 7

Efficiency gain of dynamical control plane

1 4

Multicast performance

9 10

Connectionless vs. connection oriented

3 11 12

CapEx cost benchmarking

2

Location dependent performance

8

OpEx benchmarking

13 Performance benchmarking Cost benchmarking

1.

Dynamical control plane efficiency gain

3.

MSTP vs. ELS provisioning efficiency

4.

Provisioning (ELS validation test)

5.

Connectivity control & OAM experimentation

6.

Recovery experimentation

7.

QoS experimentation

8.

Edge functionality

9.

Multicast

10.

Performance comparison of p2p- vs. p2mp-enabled networks

11.

CL and CO Ethernet TE optimizations

12.

CL and CO oriented Ethernet TE optimizations (2)

Performance and Cost benchmarking

SLIDE 13

13 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

ELS forwarding and control components

ELS switch

ELS FIB BFD component

ELS Fwdíng GMPLS CP

S-VID processing Socket interface with Dragon GMPLS CP Frame classification and parsing RSVP-TE daemon OSPF-TE daemon Path Comp daemon Socket interface with Dragon GMPLS CP

SLIDE 14

14 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

Performance benchmarking: (main) conclusions

1. Positive impact in resource gain with dynamic control plane - typical gain ~25%,

impact increasing with traffic variability

2. MSTP enhanced with off-line TE vs on-line dynamic TE gain from 5 to 20% (depending
n topology)
3. ELS enables linear throughput growth with efficient node CPU usage (e.g. less than

50% for rate < 1Gbps)

4. ELS data paths set-up time is proportional with the hop length and compliant with

classical SLAs (in terms of loss, jitter and delay)

5. Using BFD link/ELS data path failure detection: 12-30 ms (<50ms) with acceptable

resource consumption (CPU, BW) - data path segment recovery in 13 ms (<50 ms) even

n large topologies
6. Multipath routing allows reducing the required link capacity especially for resilience

~15% (gain saturates in case of more than two separate paths)

7. P2MP results in bandwidth gain (8% to 18% on the average link load and 18% to 28% on

the maximum link load) Note: bandwidth gains traded off by the additional P2MP states generated for the specific transport of multicast traffic

SLIDE 15

15 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER Objectives:

CapEx scaling for (future) changes in traffic, topology and OAM Qualitative influences on OpEx

Tools:

Analytical modeling Dimensioning tool Other performance experiment results

Conclusions:

Status quo for CPU of forwarding plane and memory Decrease for CPU of control plane Increase

for OAM (OPEX relevant) no real multicast support no fully dynamic control plane with topology increase

Cost Benchmarking

SLIDE 16

16 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

Current situation

< 5,5 km < 50 km < 100 km < O(100 km)

O(10) nodes

Large CO Regional POP

ATM /FR - SDH Ethernet Bridging

Metro Aggr. Switch

Core Aggregation Metro Access First Mile

< 10 nodes

IP edge routers Customer Premises IP Access router ETH Access DSLAM SP1..i-1 SPi…n Internet

IP/MPLS over SDH (NMS)

Core Aggr. Switch SDH SDH IP Core routers

NAP xDSL xDSL xDSL

SLIDE 17

17 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

Tomorrow’s situation

< 5,5 km < 50 km < 100 km < O(100 km)

O(10) nodes

Large CO Regional POP

Carrier Ethernet aggregation

Ethernet metro switch

Core Aggregation Metro Access First Mile

< 10 nodes

IP edge routers Customer Premises IP Access router IP access DSLAM SP1..i-1 SPi…n Internet Ethernet core switch 100GbE 100GbE

IP over Carrier Ethernet aggregation

xDSL NAP xDSL xDSL

ETH ETH ETH ETH

SLIDE 18

18 | TIGER - Euroview 2008 (Wurzburg) - July 21-22, 2008

TIGER

Main research directions TIGER demonstrated

1. Need for dynamic control components (in line with increasing interest in the

industry and efforts in standardization bodies)

2. Suitability of label swapped technologies (ELS and MTN)

+ Emulation platform and simulation tools are available for exploitation in future research Systematic cost modelling to assess: [functionality x performance] vs system cost equation [operational processes] vs network cost equation → still open question “IP” vs “Carrier Ethernet” : CAPEX & OPEX gain wrt functional and performance, and operational objectives ? OAM tradeoffs --> performance & design rules ? Is carrier Ethernet as designed for metro / aggregation applicable as is for core networks ?

Revenues ?