cisco asr 9000 system architectured2zmdbbm9feqrf.cloudfront.net/2015/usa/pdf/brkarc-2003.pdf ·...

Cisco ASR 9000 System Architecture

BRKARC-2003

Javed Asghar, Technical Marketing Architect - Speaker

Dennis Cai, Distinguished Engineer, Technical Marketing – Question Manager

CCIE #6621, R&S, Security

Swiss Army Knife Built for Edge Routing WorldCisco ASR9000 Market Roles

Carrier Ethernet

Mobile Backhaul

Multiservice Edge

Broadband

Gateway

DC gateway

Web/OTT

Large Enterprise

WAN

Cable/MSO

1. High-End Aggregation &

Transport

1. Mobile Backhaul

2. L2/Metro Aggregation

3. CMTS Aggregation

4. Video Distribution &

Services

2. DC Gateway Router

1. DC Interconnect

2. DC WAN Edge

3. WEB/OTT

3. Services Router

1. Business Services

2. Residential Broadband

3. Converged Edge/Core

4. Enterprise WAN

Other ASR9000 or Cisco IOS XR Sessions• … you might be interested in

• BRKSPG-2904 - ASR-9000/IOS-XR Understanding forwarding, troubleshooting the system and XR operations

• TECSPG-3001: Advanced - ASR 9000 Operation and Troubleshooting

• BRKSPG-2202: Deploying Carrier Ethernet Services on ASR9000

• BRKARC-2024: The Cisco ASR9000 nV Technology and Deployment

• BRKMPL-2333: E-VPN & PBB-EVPN: the Next Generation of MPLS-based L2VPN

• BRKARC-3003: ASR 9000 New Scale Features - FlexibleCLI(Configuration Groups) & Scale ACL's

• BRKSPG-3334: Advanced CG NAT44 and IOS XR Deployment Experience

Agenda• ASR9000 Hardware Overview

• Line Card System Architecture

• Typhoon, Trident, SIP-700, VSM

• Tomahawk

• Interface QoS Capability

• Switch Fabric Architecture• Bandwidth and Redundancy Overview

• Fabric QoS – Virtual Output Queuing

• Packet Flow and Control Plane Architecture

• Unicast

• Multicast

• L2

• IOS-XR Overview

• 32-bit and 64-bit OS

• IOS-XRv 9000 Virtual Forwarder

• Netconf/Yang

ASR9000 Hardware Overview

Compact & Powerful

Access/AggregationFlexible Service Edge

High Density Service Edge

and Core

• Small footprint with full IOS-XR feature capabilities for distributed environments (BNG, Pre-agg etc.)

• Optimized for ESE and MSE with high M-D scale for medium to large sites

• Scalable, ultra high density service routers for large, high-growth sites

Fixed 2RU

120 Gbps

2 LC/2RU

2.4 Tbps

8 LC/21RU

6.4 Tbps

10 LC/30RU

30 Tbps

20 LC/44RU

60T Tbps

4 LC/10RU

3.2 Tbps

ASR 9904

ASR 9001 / 9001-S

ASR 9006

ASR 9010

ASR 9912

ASR 9922

MSE E-MSE Peering P/PE Mobility

ASR9k Chassis Portfolio Offers Maximum FlexibilityPhysical and Virtual

nV Satellites: ASR 9000v, ASR 901/903

CEIOS XR

8 LC/21RU

24 Tbps

ASR 9910

IOS XRv

x86

Edge Linecard Silicon Slice Evolution

FutureTomahaw

k Class

800G

Tomahawk

28nm

240 Gbps

Tigershark

28nm

200 Gbps

SM15

28nm

1.20 Tbps

X86

6 Core

2 Ghz

NowTyphoon

Class

360G

Typhoon

55nm

60 Gbps

Skytrain

65nm

60 Gbps

Sacramento

65nm

220 Gbps

PowerPC

Quad Core

1.5 Ghz

Trident

90nm

15 Gbps

Octopus

130nm

60 Gbps

Santa Cruz

130nm

90 Gbps

PowerPC

Dual Core

1.2 Ghz

PastTrident

Class

120G

ASR 9001 Compact Chassis

Fixed 4x10G

SFP+ ports

Fan Tray

Field Replaceable

Redundant

(AC or DC)

Power Supplies

Field Replaceable

Sub-slot 0 with MPA Sub-slot 1 with MPA

Shipping since IOS-XR 4.2.1 May 2012

Supported MPAs:

20x1GE2x10GE 4x10GE1x40GE

Side-to-Side airflow

2RU

Front-to-back air flow with air flow baffles, 4RU, require V2 fan

ASR-9001 System Architecture Overview

MPAs

2,4x10GE

20xGE

1x40GE

SFP+ 10GE

SFP+ 10GE

SFP+ 10GE

SFP+ 10GE

Typhoon

FIA

FIA

Typhoon

Sw

itch

Fa

bric

AS

IC

RP

CPU

“RP CPU” and “linecard CPU” – same arch. as the larger systems

uses a single crossbar ASIC (just due to smaller bandwidth requirements)

MPAs

2,4x10GE

20xGE

1x40GE

On-board

2x10 SFP+

portsLC

CPUInternal

EOBC

ASR 9001-S Compact Chassis

Sub-slot 0 with MPA Sub-slot 1 with MPA

Pay As You Grow• Low entry cost

• SW License upgradable to full 9001

60G bandwidth are disabled by software. SW license to enable it

Shipping since IOS-XR 4.3.1 May 2013

Supported MPAs:

20x1GE2x10GE 4x10GE1x40GE

Side-to-Side airflow

2RU

Front-to-back air flow with air flow baffles, 4RU, require V2 fan

ASR-9001S System Architecture Overview

MPAs

2,4x10GE

20xGE

1x40GE

SFP+ 10GE

SFP+ 10GE

SFP+ 10GE

SFP+ 10GE

Typhoon

FIA

FIA

Typhoon

Sw

itch

Fa

bric

AS

IC

RP

CPU

“RP CPU” and “linecard CPU” – same arch. as the larger systems

uses a single crossbar ASIC (just due to smaller bandwidth requirements)

MPAs

2,4x10GE

20xGE

1x40GE

On-board

2x10 SFP+

portsLC

CPUInternal

EOBC

Disabled by Default

Upgradable via License

Cisco ASR 9006 OverviewFeature Description

Total Capacity 3.68T

Capacity per Slot 920G

Slots 6 slots - 4 Line Cards and 2 RSPs

Rack size 10RU

Power 1 Power Shelf, 4 Power Modules

2.1 KW DC / 3.0 KW AC supplies

Fan: Side to Side Airflow

Optional Baffle for Front-to-Back Airflow

2 Fan Trays, FRU

RSPs Integrated Fabric, 1+1 Redundancy

Line cards Tomahawk

Typhoon

VSM

SIP700 & SPAs

Side-to-back airflow, 10 RU

Front-to-back air flow with air flow baffles, 13RU, vertical


Total Capacity 7.36T

Capacity per Slot 920G


Rack size 21RU

Power: 2 Power Trays



Fan: Front to Back Airflow

2 Fan Trays, FRU


Line cards Tomahawk

Typhoon

VSM

SIP700 & SPAs


Total Capacity 6T

Capacity per Slot 3T


Rack size 6RU

Power 1 Power Trays, 4 Power Modules


Fan Side to Side Airflow, Front-to-Back Optional

1 Fan Tray, FRU


Line cards Tomahawk

Typhoon

SIP700, VSM

SW XR 5.1.0 – September 2013

Front-to-back air flow with air flow baffles, 10RU

Cisco ASR 9912 Overview

Features Description

Total Capacity 30T


Slots 10 slot chassis

Rack Size 30 RU

Power 4 Power Trays



Fan 2 Fan Trays

Front to back airflow

RP 1+1 RP redundancy

Fabric (SFC) 6+1 fabric redundancy

SW XR 4.3.2 – Shipping


Features Description

Total Capacity 60T


Slots 20 Line cards, 2 RP, 7 SFC

Rack Size 44 RU (Full Rack)

Power 4 Power Trays



Fan 4 Fan Trays

Front to back airflow

RP 1+1 RP redundancy

Fabric (SFC) 6+1 fabric redundancy

Line cards Tomahawk

Typhoon, VSM


Feature Description

Total Capacity 24T



Rack size 21RU

Power: 2 Power Trays

4.4 KW DC supplies

6.0 KW AC supplies

Fan: Front to Back Airflow

2 Fan Trays, FRU


Fabric Cards 5 Fabric cards on rear of chassis for additional capacity

230G per FC at FCS

Up to 6+1 Redundancy using RSP’s integrated fabric

Line cards Tomahawk

Typhoon

VSM

SIP700

Target IOS XR 6.1

Q1CY16

Note: This information may change until FCS

Cisco ASR 9910 Details

Target IOS XR 6.1

Q1CY16

Greater Capacity, Greater Flexibility:

Start with 2 RSPs, then add fabric

cards to scale beyond 460G per slot

ASR 9910 comes prepared with

sufficient power and cooling to

support high density 100G line cards.

Available with IOS XR 6.1 with the

same feature parity as the rest of the

ASR 9000 family.

5 Fabric Cards

2 Fan Trays

2 RSPs

8 Line Card Slots

2 Power Trays

This information may change until FCS

ASR-9910 Mid-plane ArchitectureNew Mid-plane Architecture for Greater Flexibility

M

I

D

P

L

A

N

E

Fabric card 0 – 230G

RSP0 – 230G

RSP1 – 230G

FT0

FT1

Power Tray 0

LC0

Fabric

1G/10G/40G/100G

Power Tray 1

LC71G/10G/40G/100G

Fabric card 4 – 230G

Control

Line Side

ASR 9000 Route Switch Processor Common for ASR 9904, ASR 9010, and ASR 9010

Common internal HW with RP for feature parity on IOS XR

Integrated Multi-Stage Switch Fabric

TR and SE Memory options

Time and Synchronization Support

RPS440 RSP880

Availability Q1CY12 Q1CY15

Processor Four Cores - 2.1GHz Eight Cores - 2.2GHz

NPU Bandwidth 60G 240G

Fabric Planes 5 7

Fabric Capacity 440G 880G

Memory 6G for TR

12G for SE

16G for TR

32G for SE

SSD 2x 16GB Slim SATA 2x 32GB Slim SATA

LC Support Typhoon/Trident Tomahawk/Typhoon

ASR 9904

ASR 9006

ASR 9010

ASR 9900 Route Processor Common for ASR 9912 and ASR 9922

Built for massive control place scale

Ultra High Speed Control Plane with Multi-Core Intel CPU

Huge Scale through High Memory options

Time and Synchronization Support

RP1 RP2


Processor Four Cores

2.1GHz

Eight Cores

2.2GHz

NPU Bandwidth 60G 240G

Fabric Planes 5 7

Memory 6G for TR

12G for SE

16G for TR

32G for SE

SSD 2x 16GB Slim SATA 2x 32GB Slim SATA

LC Support Typhoon Tomahawk/Typhoon

ASR 9912

ASR 9922

9904/9006/9010 RSP440

9922/9912 RP1

RSP8809922/9912

RP2

2nd Gen RP and Fabric ASIC 3rd Gen RP and Fabric ASIC

ProcessorsIntel x86

4 Core 2.27 GHz

Intel x86

4 Core 2.27 GHz

Intel x86 (Ivy Bridge EP)

8 Core 2GHz

Intel x86 (Ivy Bridge EP)

8 Core 2GHz

RAMRSP440-TR: 6GB

RSP440-SE: 12GB

-TR: 6GB

-SE: 12GB

-TR: 16GB

-SE: 32GB

-TR: 16GB

-SE: 32GB

SSD 2x 16GB Slim SATA 2x 16GB Slim SATA 2x 32GB Slim SATA 2x 32GB Slim SATA

nV EOBCports

2 x 1G/10G SFP+ 2 x 1G/10G SFP+ 4 x 1/10G SFP+ 4 x 1/10G SFP+

Punt BW 10GE 10GE 40GE 40GE

Switch fabric bandwidth

220G + 220G (9006/9010)

385G + 385G (9904)

(fabric integrated on RSP)

660G+110G

(separated fabric card)

450G + 450G (9006/9010)

800G + 800G (9904)

1.61Tb + 230G

(separated fabric card)

Route Switch Processors and Route ProcessorsRSP used in ASR9904/9006/9010, RP in ASR9922/9912

23

ASR 9900 - Switch Fabric Cards

In-Service

Upgrade

Common for ASR 9912 and ASR 9922

7 Fabric Card Slots

Decoupled, multi-stage switch fabric hardware

True HW separation between control and data plane

Add bandwidth per slot easily & independently

Similar architecture to CRS

SFC110 SFC2


Fabric Capacity

per SFC

110G 230G

Fabric Capacity

Per Line Card Slot

660G N+1

770G N+0

1.38T N+1

1.61T N+0

Fabric

Redundancy

N+1 N+1

LC Support Typhoon Tomahawk/Typhoon

ASR 9912

ASR 9922

New ASR-9922 and ASR-9006 V2 Fans

ASR 9922-FAN-V2

IOS XR 5.2.2

ASR 9006-FAN-V2

IOS XR 5.3.0

V2 Fan provides higher cooling capacity for ultra high density cards.

Motor and blade shape optimized to produce higher CFM.

New material capable of dissipating more heat

In service upgrade

PAYG Power N+1 Redundancy for DC

N+N Redundancy for AC

Typhoon and Tomahawk LCs supported with V2

Plan for the future beyond 1T per slot with V3

Version AC DC Chassis

V2 3.0KW 2.1KW ASR9006, ASR9010,

ASR9904, ASR9912, ASR9922

V3 6.0KW 4.4KW ASR9010, ASR9912, ASR9922

Line Card Used BW Consumption at 27C Consumption at 40C

Typhoon 360G / Slot 2.8 Watts/G 3.5 Watts/G

Tomahawk 1T / Slot 1.6 Watts/G 1.9 Watts/G

Benefit of Power Reductions for 100G4 Tbps: Tomahawk vs. Typhoon in Provider-owned European Data Center

4x 100G Density vs Typhoon

$3460 per port per year power savings

$1,384,000 savings over 10 years

Notes: 9922 with 20 2x100G vs. 9912 with 5 8x100G 8-year facility

amortization with 1.7 PUE, 40C max, France power (USD $.19 per

kWh) N:N Power redundancy

Year

Po

we

r Co

st

Line Card System Architecture1. Typhoon, Trident, SIP-700, VSM2. Tomahawk3. Inteface QoS

Modular SPA Linecard• 20Gbps, feature ritch, high scale, low speed Interfaces

SIP-700

• Distributed Control and

Data Plane

• 20Gbits, 4 SPA Bays

• L3 i/f, route, session

protocol – scaled for MSE

needs

Scalability

• Flexible uCode Architectue

for Feature Richness

• L2 + L3 ServicesL FR, PPP,

HDLC, MLPPP, LFI

• L3VPN, MPLS, Netflow,

6PE/6VPE

Powerful & Flexible QFP

Processor

• IC-Stateful Switch Over

Capability

• MR-APS

• IOS-XR base for high

scale and Reliability

High Availability

• ChOC-3/12/48 (STM1/4/16)

• POS: OC3/STM1, OC12/STM4,

OC-48/STM16, OC192/STM64

• ChT1/E1, ChT3/E3, CEoPs, ATM

SPA Support

SPAs

• 128k Queues

• 128k Policers

• H-QoS

• Color Policing

Quality of Service

ASR 9000 Ethernet Line Card Overview

A9K-40G A9K-4T A9K-8T/4 A9K-2T20G A9K-8T A9K-16T/8

A9K-36x10GE

A9K-2x100GE

(A9K-1x100G)

A9K-24x10GE

A9K-MOD80

A9K-MOD160

MPAs20x1GE2x10GE 4x10GE8x10GE1x40GE 2x40GE

-L, -B, -E

-TR, -SE

First-generation LC

Trident NPU:

15Gbps, ~15Mpps,

bi-directional

Second-gen LC

Typhoon NPU:

60Gbps, ~45Mpps,

bi-directional

-L: low queue, -B: Medium queue, -E: Large queue, -TR: transport optimized, -SE: Service edge optimized

• Micro-CPU based forwarding chip

• Feature flexibility, future proven

• Programmable forwarding tables

• High Performance

• Line rate performance on all line cards

• End-to-End Internal System QoS

• Efficient multicast replication

• High Control Plane Scale

• 4M IPv4 or 2M IPv6 FIB per line card

• 2M MACs learned in hardware

24x10GE

2x100GE

ASR 9000 80-360G “Typhoon Class” LinecardsHyper Intelligence & Service Scalability

36x10GE

Modular 80G & 160G

Network Processor Architecture Details

• TCAM: VLAN tag, QoS and ACL classification

• Stats memory: interface statistics, forwarding statistics etc

• Frame memory: buffer, Queues

• Lookup Memory: forwarding tables, FIB, MAC, ADJ

• TR/SE

• Different TCAM/frame/stats memory size for different per-LC QoS, ACL, logical interface scale

• Same lookup memory for same system wide scale mixing different variation of LCs doesn’t impact system wide scale

-

STATS MEMORY

FRAME MEMORYLOOKUP

MEMORY TCAM

FIB MAC

NPU Complex

Forwarding chip (multi core)

TR and SE has different

memory sizeTR and SE has same

memory size

-TR: transport optimized, -SE: Service edge optimized

NP

Switch Fabric

3x10GE

SFP +

3x10GE

SFP +

3x10GE

SFP +

3x10GE

SFP +

3x10GE

SFP +

3x10GE

SFP +

3x10GE

SFP +

3x10GE

SFP +

FIA

FIA

FIA

FIA

Sw

itch

Fab

ric A

SIC

CPU

PuntFPGA FIA

CPU

Switch Fabric

RP

LC1

3x10GE

SFP +

3x10GE

SFP +

NP

NP

3x10GE

SFP +

3x10GE

SFP +

NP

NP

3x10GE

SFP +

3x10GE

SFP +

NP

NP

3x10GE

SFP +

3x10GE

SFP +

NP

NPFIA

FIA

FIA

FIA

Sw

itch

Fa

bric

AS

IC

CPULC2

1 2

2

NP learn MAC address in hardware (around

4M pps)

NP flood MAC notification (data plane)

message to all other NPs in the system to sync

up the MAC address system-wide. MAC

notification and MAC sync are all done in

hardware

1

2

NP

NP

NP

NP

NP

NP

NP

Hardware based MAC learning: ~4Mpps/NP

MAC Learning and Sync

Data packet

33

Typhoon LC: 24x10G Ports Architecture

3x 10G

FIA

30G3x10GE

SFP +

3x10GE

SFP +

Typhoon

Typhoon

3x 10G

30G

FIA

3x 10G

30G3x10GE

SFP +

3x10GE

SFP +

Typhoon

Typhoon

3x 10G

30G

FIA

3x 10G

30G3x10GE

SFP +

3x10GE

SFP +

Typhoon

Typhoon

3x 10G

30G

FIA

3x 10G

30G3x10GE

SFP +

3x10GE

SFP +

Typhoon

Typhoon

3x 10G

30G

Line cardlocal

Fabric

Complex

(NextGen

S/F ASIC)

RSP 440 Switch

Fabric

RSP 0

RSP 1

8x55G

60G

90G

60G

90G

60G

90G

60G

90G

No Congestion Point

Non-Blocking everywhere

Typhoon NPU BW and Performance1. 120G and 90Mpps Uni-directional

2. 60G and 45Mpps full-duplex (each direction ingress/egress)

Typhoon Line Card Architectures36x10G Port LC 2x100G Port LC

MOD80 LC MOD160 LC

Metric Trident Typhoon

FIB Routes (v4/v6) 1.3M/650K* 4M/2M

Multicast FIB 32K 128K

MAC Addresses 512K* 2M

L3 VRFs 4K 8K

Bridge Domains / VFI 8K 64K

PW 64K 128K

L3 Subif / LC 4K 8K (TR)

20K (SE)

L2 Interfaces (EFPs) / LC 4K (-L)

16K (-B)

32K (-E)

16K (TR)

64K (SE)

MPLS labels 256K 1M

Queue Scale (per NP) 64K egress + 32K ingress 192K egress + 64K ingress

Policer scale (per NP) 32K/64K (-L,-B/-E) 32K/256K (-TR/-SE)

Trident vs. Typhoon– Major Feature Scale

* It require scale profile configuration to reach maximum FIB or MAC on the Trident line card. On Typhoon, FIB and MAC has dedicated memory

ASR9k Full-Mesh IPv4 PerformanceASR9010,

2x RSP-440

8x Typhoon 24x10G = 192 ports

Full mesh of flows = ~72,000 flows

(36k Full-duplex Flows)

.

.

.

.

.

.

.

.

IXIA

Tx/RxIXIA

Tx/Rx

ASR9k

A9k-24x10GE – L3 Multicast Scale Performance Setup and Profile

ASR 9010

A9k-2

4x1

0G

E

.

.

.

.

.A9k-2

4x1

0G

E

port 0

port 1

port 21

port 22

port 23

IXIA Multicast Receivers

L3 sub-interface multicast

receiver facing ports





port 10

L3 sub-interface multicast

Source facing ports

IXIA Multicast Source

Scale Profile:1. Unidirectional Multicast

2. 2k (S,G) mroutes

3. 24 OIFs per (S,G) mroute

10G IGMP Joins

A9k-24x10GE – L3 Multicast Scale Throughput Performance Results

240G 240G 240G 240G 240G 240G 240G 240G 240G357M

203M

109M

56M

29M 20M 15M 7M 3M 020406080100120140160180200220240260280300320340360

0

20

40

60

80

100

120

140

160

180

200

220

240

260

64 128 256 512 1,024 1,518 2,048 4,096 9,000

Mfp

s

Gb

ps

Frame Size

A9k-24X10GE - L3 Multicast Throughput PerformanceScale: 2k (S,G), 24 OIFs per (S,G)

Agg Rx Throughput (Gbps) Agg Rx Throughput (Mfps)

ASR 9000 VSM• Data Center Compute:

• 4 x Intel 10-core x86 CPU

• 2 Typhoon NPU for hardware network processing

• 120 Gbps of Raw processing throughput

• HW Acceleration

• 40 Gbps of hardware assisted Crypto throughput

• Hardware assist for Reg-Ex matching

• Virtualization Hypervisor (KVM)

• Service VM life cycle management integrated into IOS-XR

OS / Hypervisor

VMM

VM-4

Service-3

VM-1

Service-1VM-3

Service-4

VM-2

Service-2

CGN

Shipping

IPSec

ShippingSecGW

Q2/Q3 CY15

Firewall

On Radar

Anti-DDOS*

Q1,CY15

3rd Party Apps*

Virtualized Services Module (VSM) Overview

DDR3DIMM

DataPath

Switch

Virtualized Services Sub- Module Router Infrastructure Sub-Module

10

GE/

FCo

ESF

P+

Linecardlocal Fabric

Complex

RSP/RP Switch Fabric

RSP/RP 0

RSP /RP1

Crypto

DDR3DIMM

Crypto

DDR3DIMM Crypto

DDR3DIMM

Crypto

Typhoon NPU

Typhoon NPU

VSM Details

Tomahawk 8x100GE CPAK Line Card

Tomahawk 4x100GE CPAK Line Card

LAN Only Shipping 5.3.0

LAN/WAN/OTN April 2015

(5.3.1)

LAN/WAN/OTN April 2015

(5.3.1)

New Tomahawk LCs

CPAK 100 GE LR4CPAK 100 GE ER4 CPAK 10x10-LR

CPAK 100 GE SR10CPAK Options

100 GE LR4 10x10 GE 2x40 GE 100 GE SR10

Investment Protection

Start with 10 GE and upgrade to 100 GE in the future

Flex 100G CPAK

Tomahawk LC: 8x100GE Architecture

XB

AR

(S

M1

5)

Ce

ntra

l XB

AR

Ivy Bridge CPU Complex

TomahawkNP

TigersharkFIA

PHY

TomahawkNP

TigersharkFIA

PHY

TomahawkNP

TigersharkFIA

PHY

TomahawkNP

TigersharkFIA

PHY

CPAK:

100G, 40G, 10G

L2/L3/L4 lookups, all VPN types, all

feature processing, mcast replication,

QoS, ACL, etc …

Slice Based Architecture

Macsec Suite B+, G.709, OTN,

Clocking

VoQ buffering, Fabric credits, mcast

hashing, scheduler for fabric and

egress port

240G

240G

240G

240G

240G

240G

240G

240G

FPOE, Auto-Spread,

DWRR, RBH, replication

Per Slice Power Management: (100-200W Power Savings)

PE1(admin-config)# hw-module power disable slice [0-3] location

Metric Tomahawk-SE Scale Typhoon –SE Scale

MPLS Labels 1M 1M

MAC Addresses2M @ FCS

6M - future2M

FIB Routes (v4/v6) – Search

Memory

4M(v4) / 2M(v6) - XR

10M(v4) / 5M(v6) possible in

future release

4M(v4+v6)

VRF 16K 8K/LC

Bridge Domains 256K/LC 64K/LC

TCAM TCAM (80Mb) TCAM (40Mb)

Packet Buffer12GB/NPU

200ms

2GB/NPU

100ms

EFPs 256K/LC (64K/NP) 64K/LC

L3 Subif (incl. BNG) 128K (64K/NP) 20k/LC (sys)

Egress Queues 1M/NPU (4M for 8x100GE) 256k/NPU

Policers 512k/NPU 256k

-Note: Actual available LC/NPU capacity depends on software release and subject to change!

Tomahawk Vs Typhoon Hardware Capability

Tomahawk Line Card CPU

CPU Subsystem:

Intel IVY Bridge EN 6 cores @ ~2GHz with 3 DIMMs

Integrated acceleration engine for Crypto, Pattern Matching and Compression

1x32GB SSD

HW Parameter Typhoon LC CPU Tomahawk-SE Linecard

ProcessorP4040

4 Core 1.5GHz

Ivy Bridge EN

6 Core 2GHz

LC CPU Memory 4GB 24 GB

Cache L1, L2, L3

L1: 32KB Instructions,

L2: 256KB

L3: 2.5MB per core

• MPA’s Supported:2x100GE CPAK. 1x100GE CPAK. 20x10GE SFP+All Typhoon MPA’

• Flexibility to use CPAK 10G/40G/100G optics on 100G MPAs

• 2 Flavors for FlexibilityMOD400 has 2 Tomahawk ASICS (FCS – August 2015)MOD200 has 1 Tomahawk ASIC (FCS – Oct 2015)

MPA #1(2x100G)

MPA #2(20x10G)

Mo

d4

00

LC

2 x 100GCPAKs

20 x 10G SFP+

MOD200Support Matrix

Comb 1 Comb 2 Comb 3

EP0 2x100G-MPA 20x10G-MPA

1x100G orany Typhoon MPAs

EP1 None None 1x100G or any Typhoon MPAs

MOD400Support Matrix

Comb 1 Comb 2 Combo 3 Comb 4

EP0 2x100G-MPA 20x10G-MPA 2x100G-MPA 1x100G or anyTyphoon MPAs

EP1 2x100G-MPA 20x10G-MPA 20x10G-MPA 2x100G-MPA OR20x10G-MPA

Tomahawk Modular LCs – Mod400 & Mod200

Tomahawk Line Card Architectures8x100G/80x10G Port LC 4x100G/40x10G Port LC

MOD400 LC MOD200 LC

12x100G Tomahawk LC “Skyhammer”

High Level Differences:Tomahawk 8x100G and 12x100G

8x100GE “Octane” 12x100GE “Skyhammer”

8-port 100GE with 4 NPU slices 12-port 100GE with 6 NPU slices

CPAK Optics QSFP28 Optics

5-Fabrics (7-Fabric in roadmap for Nov/Dec 2015 FCS) 7-Fabric card

Compatible with all 99xx-series and 90xx-series chassis Support only for 99xx-series chassis

Full L3VPN/L2VPN feature support Only L3 features in Ph-1, L2 support in Ph-2

Has external TCAM for High Qos/ACL scale Has no external TCAM . Will use only 5Mb internal TCAM

Total TCAM entries: 192K Total TCAM entries: 32K

32K sub-interfaces 1K sub-interfaces

4M V4 Route, 2M V6 routes Under discussion

Tomahawk TCAM Optimized ScaleMetric Tomahawk

-TR Scale

Tomahawk

-SE Scale

Skyhammer

Scale

MPLS Labels 1M

Addresses 2M (6m Future)

FIB Routes (v4/v6) – Search

Memory

4M(v4) / 2M(v6) – XR (10m/5m Future)

Mroute/MFIB (v4/v6) 128k/32k (256k/64k Future)

VRF 8K (16k Future)

Bridge Domains 64K

TCAM (acl space v4/v6) TCAM – 1/4 SE TCAM (80Mbit) Internal TCAM (5Mbit)

Packet Buffer Packet Buffer – 100ms

(6G/NPU)

Packet Buffer – 200ms

(12G/NPU)

Packet Buffer – 100ms

EFPs 16K/LC 128K/LC (64K/NP) N/A

L3 Subif (incl. BNG) 8K 128K (64K/NP) 1K

IP/PPP/LAC subscriber

sessions per LC

16K 256K (64K/NP) N/A

Egress Queues 8 Queues / port + nV Sat Q’s 1M/NPU (4M for 8x100GE!) 8 Queues / port

Policers 32K/NPU 512K/NPU

QOS/ACL (v4/v6) 16k v4 or 4k v6 ACEs/LC 98k v4 / 16k v6 ACEs 24K v4/1.5K v6

ASR9k Optical Solution Before IPoDWDM LCASR9k + NCS2k Optical Shelf

Low Cost

Interconnect

ASR9000 NCS2k

100GE Linecard Coherent Transponder

SR10

CFP/CPAK

transceiver

SR10

CXP/CPAK

Integrated nV Optical System

• Tomahawk based Linecard

• Feature and scale parity with other –TR and –SE Tomahawk cards

• 2xCFP2 based DWDM ports (100G, 200G)

• BPSK, QPSK, 16QAM modulation options

• 96 channels, ITU-T 50GHz spacing

• FlexSpectrum support

• HD FEC, SD FEC (3000+ km w/o regen)

• 20x10GE SFPP ports (SR, LR, ZR, CWDM, DWDM)

• Flexible port options up to 400 Gbps total capacity

• 2 x 200G DWDM (CFP2) or

• 2 x 100G DWDM (CFP2) + 20 x 10G (SFP+) or

• 1 x 100G + 1x200G DWDM (CFP2) + 10 x 10G (SFP+)

• OTN and pre-FEC FRR

• Target FCS: mid-CY15 (5.3.2) for 100G DWDM

• 10G Gray Ports and 200G DWDM in a future release

FCS Mid-CY15

400G IPoDWDM LC Overview (Tomahawk Based LC)

SM15

FIA

VOQ

Tomahaw

k

NPMEM

X240EtnaCoherent

CFP2

200G capable

10GE SFP+

10GE SFP+

#1

#20

HD-FEC

FPGA

TCAM

SerDes

Mux

FIA

VOQ

Tomahaw

k

NPMEM

X240EtnaCoherent

CFP2

200G capable

HD-FEC

FPGA

TCAM

X240

Ivy Bridge

CPU Complex

400G IPoDWDM LC Specific HW components

400Gbs IPoDWDM LC Internals

SM15

FIA

VOQ

Tomahaw

k

NPMEM

X240EtnaCoherent

CFP2

200G capable

10GE SFP+

10GE SFP+

#1

#20

HD-FEC

FPGA

TCAM

SerDes

Mux

FIA

VOQ

Tomahaw

k

NPMEM

X240EtnaCoherent

CFP2

200G capable

HD-FEC

FPGA

TCAM

X240

Ivy Bridge

CPU Complex

#0

#1

Tomahawk Per Slice MACSEC PHY Capability • MACSEC Security Standards Compliant with:

• IEEE 802.1EA-2006

• IEEE 802.1AEbn- 2011 (256-bit key)

• IEEE 802.1AEbw-2013 (extended packet numbering)

• Security Suites Supported:

• AES-GCM-128, 128-bit key (32 bits)

• AES-GCM-256, 256-bit key (32 bits)

• AES-GCM-XPN-128, provides extended packet number counter (64 bits)

• AES-GCM-XPN-256, provides extended packet number counter (64 bits)

• Unique Security Attributes Per Security Association (SA):

• 10G port = 32 SA

• 40G port = 128 SA

• 100G port = 256 SA

• Per Slice Port Combination Supported (CPAK)

• 2x100G, 20x10G, 4x40G, 1x100G + 10x10G, 2x40G + 10x10G, 2x40G + 1x100G

• All Tomahawk LC variations support MACSEC

• 8x100G, 4x100G, MOD-400, MOD-200

ASR9k MACSEC Phase 1: XR 5.4.0 ReleaseSP/CE/DCI/Enterprise Usecases

Usecase #2: Link MACSEC over LAG members

MACSEC Links

MACSEC

on LAGMember link

InheritanceCE CEPEPE P

Usecase #1: Link MACSEC in MPLS/IP Topology

MACSEC Links

ASR9k

DCI/CE DCI/CE

Usecase #3 CE Port Mode MACSEC over L2VPN

MKA

L2VPN

CE/WAN

MACSEC Links

port

modeport

mode

ASR9k ASR9k

DCI/CE DCI/CE

Usecase #4 VLAN Clear Tags MACSEC over L2VPN

MKA

L2VPN

CE/WAN

MACSEC Links

vlan

clear-tags

vlan

clear-tags

ASR9k ASR9k

Secure L2VPN as a Service: PW/EVPN/PBB (any L2) Circuit Encryption using MACSEC

PHY NPUFabric

1

0

G

1

0

G

1

0

G

Port 1

Port 2

Port 3

Clear Frame

EFP1

EFP2

EFP3

Phy Secure Channel

Loopback + Clear Tags

(Cisco IPR)

DA SA VLA

N

PAYLOA

D

FCS

Frame Encryption on Port 2 inside PHY

DA SA VLA

N

SEGTA

G

PAYLOA

D

ICV FCS

Clear Frame

DA SA VLA

N

PAYLOA

D

FCS

Frame bypass Port 3 MACSEC inside PHYEncrypted EoMPLS PW

VC

Label

DA SA VLA

N

SEGTA

G

PAYLOA

D

ICV FCS

DA SA VLA

N

SEGTA

G

PAYLOA

D

ICV FCS

Tomahawk MACSEC

Raw Performance

• Tomahawk MACSEC

• Raw Scale

AES-GCN-256-bit Raw Performance (full-duplex)

Per LC Slice 200Gbps

Per LC Slot 800Gbps

Per Chassis ASR9006 = 3.2Tbps

ASR9010 = 6.4Tbps

ASR9904 = 1.6Tbps

ASR9012 = 8Tbps

ASR9922 = 16Tbps

AES-GCN-256-bit Raw Scale

Total MACSEC Ports

Per System

10G = 1,600

40G = 320

100G = 160

Total MACSEC SAs

Per System

10G Tx/Rx SAs = 51,200

40G Tx/Rx SAs = 40,960

100G Tx/Rx SAs = 40,960

• Generally target >3.3x performance of Typhoon system

• Cust benchmark: 2x 100GE/Tomahawk NPU vs 6x 10GE/Typhoon NPU

Customer Profile Tomahawk Performance

Customer

ProfileApplication Profile

Tomahawk

LR Frame

Typhoon LR Frame

@6x10G

Performance Incr

Ratio

Web/OTT 1

LSR: top label swap + Bundle + TE Tunnel + iMarking +

eMarking/Queue 160B IP

@126mpps

[email protected] 3.8x

LER: MPLS imp + ieQoS + iNF + ieBundle + RSVP

TE/[email protected] 5.5x

Web/OTT 2

LSR: label swap + Bundle + TE Tunnel + iMarking +

eMarking/queuing 160B IP

@126mpps

[email protected] 3.8x

LER: IP + Bundle +TE Tunnel + iMarking +

eMarking/queuing + iNF + [email protected] 6.3x

Tier1

ISP/Peering

1

Ipv4 & v6 recursive + in/out ACL + uRPF + in NF + iMAC

Accounting460B IP

@[email protected] 4.2x

Tier1

SP/Peering

2

IP to IP (recursive) + QoS(out) + ACL(in+out) +

Netflow(1:5K,in) + Bundle286B IP

@81.6mppsiMix LR@6x10G > 3.3x

• Generally target 3.3x performance of Typhoon system

• Benchmark set: 2x100GE/T-hawk NPU vs 6x 10GE/Typhoon NPU

Tomahawk Baseline Performance

LineCard

NPU

BW

Packet

Fwding

Capacity

IPv4 NR

+ Rx ACL

IPv6 NR +

Rx ACL

IPv4 + in NF

(1:1k)

IPv6

In Policing

+Out Shaping

L2 Xconnect VPLS + QoS

Tomahawk 240G149

mpps

128B Eth

@

169mpps

196B Eth

@

115mpps

225B Eth

@

102mpps

233B Eth

@

104mpps

123B Eth @

174.6mpps

449B Eth

@

53.3mpps

Typhoon 60G45

mpps

178B Eth

@

37.8mpps

300B Eth

@

23.4mpps

205B Eth

@

33.4mpps

316B Eth

@

22.3mpps

148B Eth @

44.6mpps

445B Eth

@

16.1mpps

Perf Incr Ratio 4x 3.3x 4.48x 4.48x 3.07x 4.42x 3.91x 3.3x

Jan 2015XR 5.3.0

8x100GE LAN PHY*

8x100GE OTN*

4x100GE OTN

MOD400

400G IPoDWDM

12x100G

MOD200

April 2015XR 5.3.1

August 2015XR 5.3.2

Commons

Line Cards

RP2

RSP880

SFC2

Mar 2016XR 6.1.0

All Platforms

9910, 9912, 9922

90xx, 9904, 9910

Nov 2015XR 6.0.0

8x100G 7-Fab

* Oversubscribed on 9010, 9006 with Single RSP

9912, 9922

Tomahawk Roadmap

Tomahawk Line Card Port QOS Overview• 4 Priority Interface Queuing:

• PQ1, PQ2, PQ3 Strict Priority• PQ4 = Remaining Queues (CBWFQs)

• Configurable QoS policies using IOS XR MQC CLI

• QoS policy is applied to interface (physical, bundle or logical*), attachment points

– Main InterfaceMQC applied to a physical port will take effect for traffic that flows across all sub-interfaces on that

physical port will NOT coexist with MQC policy on sub-interface ** you can have either port-based or subinter-face based policy on a given physical port

– L3 sub-interface– L2 sub-interface (EFP)

• QoS policy is programmed into hardware microcode and queue ASIC on the Line card NPU

* Some logical interface could apply qos policy, for example PWHE, BVI

** it could have main interface level simple flat qos co-exist with sub-interface level H-QoS on ingress direction

Tomahawk Line Card Port QoS Overview – con’t

• 5 level hierarchy flexible queuing/scheduling support

– 5 level scheduling hierarchy:

Port groups (L0), Ports (L1), Logical Ports (L2), Subinterfaces (L3), Classes (L4)

– Egress & Ingress, shaping and policing

• Three strict priority scheduling with priority propagation

• Flexible & granular classification, and marking

–Full layer 2, full layer 3/4 IPv4, IPv6, mpls

• Dedicated queue ASIC – TM (traffic

manager) per NPU for QoS functions

• -SE and -TR LC version has different queue

buffer/memory size, different number of

queues

NPUFIA Switch

Fabric ASIC

TM

Port G

roup

L1

Port

L2

Logical Port

5 Level Hierarchy QoS Queuing Overview

L4

Classes

• L0, L1 level schedulers are automated

by TM, not user configurable

• L2, L3 and L4 can be flexibly mapped

to parent level scheduler

• Hierarchy levels used are determined

by how many nested levels a policy-

map is configured for and applied to a

given sub-interface

• Up to 16 classes per child/grandchild

level (L4)

Port

C-VLAN

C-VLAN

Class

Class

Class

Class

C-VLAN

C-VLAN

Class

Class

Class

Class

L0

PG

EFP1 (S-Vlan

or Vlans)

EFP2 (S-Vlan

or Vlans)

L3

Sub-interfaces

Queue/Scheduler Hierarchy MQC Capabilities

Priority 1

P2

P3

Normal

Pri Qs

L1

Shape (PIR,

or Port

Shaper)

L2: 2-param

Shape

BRR Weight

(Bw or BwR)

L3: 3-param

Shape

Bandwidth

BRR W.

L4: 2-param

Shape, BRR W (Bw or BwR), Priority, WRED/Q-Limit

P1

P2

Normal

Pri Qs

QoS Classification Criteria

L2 Header Fields L3 Header Fields Internal Marking

L2

Interfaces/EF

Ps

Or L3

Interfaces

Inner/outer COS,

inner/outer vlan,

DEISource/Destination MAC address*match all/match any

Outer EXPDSCP/TOSTTL, TCP flags, Source/destination L4 portsProtocolSource/Destination IPv4address*

Discard-classQos-group

Notes: - Support match all or match any

- Max 8 match statements per class, max 8 match entries per match statement

- Not all header fields can be used in one MQC policy-map, see details next

Tomahawk 3 Level Policing• Supports grand parent policing

• Conform-aware coupled between child and parent, parent and grandparent- Only 1R2C policer at grand-parent level

• Color aware policing 1R2C, MEF 2R3C, No IETF 2R3C- Color value can be one of the fields: outer CoS, Exp, Prec, QoS-group, DSCP and Discard-class

- Only one conform-color value per policer, others are considered as exceed-color

• Color-aware coupled policing- Child level: incoming packet’s color field value is used for matching a color (Parent or child remarking will not take effect for color matching)

- Parent level: the child level marking if any will be effective in matching the color at parent level

- Grand parent level: only support 1R2C without color aware

Child-Level Parent-Level Grand-Parent Level Allowed

No-policer No-policer Policer Allow

Policer No-Policer Policer Allow

No-Policer Policer Policer Allow

Policer Policer Policer Allow

Coupled Child Coupled Parent Policer Reject

Coupled Child No-Policer Coupled-Parent Reject

Policer Coupled Child Coupled-Parent Allow

Shaping, Policing Overhead Accounting

• L2 frame length is being used by default without preamble and IFG. • Same for ingress/egress • Same for all QOS actions

• QoS policy can be configured to take into account arbitrary L1 framing and L2 overhead

Preamble7

SFD1

DA6

Length/Type

2Payload46-1500

FCS4

SA6

Inter Frame Gap12

VLAN4

Switch Fabric Architecture1. Bandwidth and Redundancy Overview2. Fabric QoS – Virtual Output Queuing

CPU

FIA

NP FIA

FIA

CPU

SerDesXBAR

NP

EP0

PHY

EP1

PHY

SwitchFabric(SM15)

Cisco ASR 9000 High Level System Architecture“At a Glance”

72

Linecard

Switch Fabric

integrated on RSP

or Separated

RSP/RP

CPU

FIA

NP FIA

FIA

CPU

SerDesXBAR

NP

EP0

PHY

EP1

PHY

SwitchFabric(SM15)

Multi stage (1,2,3)

fabric operation

Scalable & flexible

Slice based data plane

Fully distributed &

Redundant system

ASR 9006/9010 Switch Fabric Overview3-Stage Non-Blocking Fabric (Separate Unicast and Multicast Crossbars)

FIAFIA

FIARSP0

Arbiter

fabric

RSP1

Arbiter

fabric

Fabric bandwidth:

8x55Gbps =440Gbps/slot with dual RSP

4x55Gbps =220Gbps/slot with single RSP

fabricFIAFIA

FIA

fabric

Fabric frame format:

Super-frame

Fabric load balancing:

Unicast is per-packet

Multicast is per-flow8x55Gbps

Stage 1 Stage 2 Stage 3

Typhoon LCTyphoon LC

Ingress LinecardEgress Linecard

RSP440

8x55Gbps

Active-Active Fabric

Unicast

Crossbar

Multicast

Crossbar

73

Ingress PortQoS

End-to-End priority (P1,P2, 2xBest-effort) propagation Unicast VOQ and back pressureUnicast and Multicast separation

Egress PortQoS

Ingress side of LC Egress side of LC

NP PHY

NP PHYFIA

CPU

NPPHY

NPPHY

FIA

CPU

Switch Fabric

1

3

4

1 2 3 4

4 Priority Egress Destination

Queues (DQs) per SFP (VQI)

virtual port, aggregated at

egress port rate

4 VOQ per each SFP virtual port in the entire system

Up to 8K VOQs per TSK FIA (vs 4k per SKT FIA)

ASR9k End-2-End System QoS Overview

2

74

ASR9k Virtual Output Queuing (VoQ) System Architecture VoQ Components (Where are they)

Egress NPU Backpressure and VoQ in ActionResult is No Head of Line Blocking (HOLB)

ASR9904 RSP880 Switch Fabric ArchitectureActive/Active 3-Stage Fabric, Scale to 1.6Tbps LCs

FIAFIA

FIA RSP880

Arbiter

RSP880

FIAFIA

FIA


Tomahawk Line CardTomahawk Linecard

Ingress Linecard Egress LinecardFabricSM15

FabricSM15

FabricSM15


Super-frame



Multicast is per-flow

Fabric bandwidth:

10x 115Gbps ~1.1Tbps/slot with dual RSP 5 x 115Gbps ~ 575Gbps/slot with single RSP

2x 5x 115Gbps ~ 1.15Tbps

2x 5x 115Gbps ~ 1.15Tbps

Arbiter

FabricSM15

Fabric bandwidth:

10x 115Gbps ~1.15Tbps/slot with dual RSP 5 x 115Gbps ~ 575Gbps/slot with single RSP

77

ASR90xx – RSP880 and Mixed LC Operation

FIAFIA

FIARSP880

Arbiter

RSP880

FIAFIA

FIA


Tomahawk Line CardTyphoon Linecard


Fabric bandwidth:

8x115Gbps ~ 900Gbps/slot with dual RSP

4x115Gbps ~ 450Gbps/slot with single RSP


Super-frame




8x115Gbps

8x55GbpsFabric bandwidth:



Fabric

Fabric

SM15

Fabric

SM15

Arbiter

FabricSM15

78

ASR90xx – RSP440 and Mixed LC Operation

FIAFIA

FIARSP0

RSP1

fabricFIAFIA

FIA

Arbiter

fabric


Tomahawk Line CardTyphoon Linecard


RSP440

Fabric bandwidth:




Super-frame




8x55Gbps

8x55GbpsFabric bandwidth:



Fabric

SM15

Arbiter

fabric

79

SFC v2

Tomahawk Line Card

FIAFIA

FIA

ASR99xx Switch Fabric Card (FC2) Overview6+1 All Active 3-Stage Fabric Planes, Scale to 1.6Tbps LCs

Tomahawk Line Card

FIAFIA

FIA

5x 2x115G (120G raw)

bi-directional

= 1.15Tbps

5x2x115G bi-directional

= 1.15Tbps

Fabric

SM15

Fabric

SM15


Super-frame

Fabric load

balancing:



Fabric bandwidth:

10x115Gbps ~ 1.15 Tbps/slot with 5x FC2

8x115Gbps ~ 920 Gbps/slot with single RSP

80

SFC v2

Tomahawk Line Card

FIAFIA

FIA

ASR9922/12 – SFC2 and Mixed Gen LCs When using 3rd Gen Fabric Cards

Typhoon Linecard

FIAFIA

FIAfabric

(5-1)x2x55G bi-directional

= 440Gbps (protected)

(5-1)x2x115G bi-directional

= 920Gbps (protected)

Fabric Lanes 6 & 7 are only used towards

3rd Gen 7-fabric plane linecards

Fabric

SM15

81

ASR9K Tomahawk Module Fabric Redundancy and Bw Allocation Summary for 8x100G LCs with RSP880s/SFCv2

Fabric

Redundancy

Per Slot

Fabric BW

Per 8x100G

LC Data BW

System

QoS/Priority

Protection

System

Dual RSP880s 920G 800G Y ASR9010

ASR9006Single RSP880 460G 400G Y

Dual RSP880s 1.15T 800G YASR9904

Single RSP880 575G 500G Y

5x SFCv2 1.15T 800G Y

ASR9922

ASR9912

4x SFCv2 920G 800G Y



Packet Flow and Control Plane Architecture1. Unicast2. Multicast3. L2

ASR9000 Fully Distributed Control Plane

84

Switch Fabric

3x10GE

SFP +

3x10GE

SFP +

Typhoo

n

NP

3x10G

E

SFP +3x10GE

SFP +

NP

NP

3x10GE

SFP +

3x10GE

SFP +

NP

NP

3x10GE

SFP +

3x10GE

SFP +

NP

NPFIA

FIA

FIA

FIA

Sw

itch

Fab

ric A

SIC

PuntFPGA FIA

Switch Fabric

RP

LPTS

LC CPU: ARP, ICMP, BFD, NetFlow,

OAM, etc

RP CPU: Routing, MPLS, IGMP, PIM,

HSRP/VRRP, etc

LPTS (local packet transport service):

control plane policing

Control packet

Punt Switch

CPU

CPU

Layer 3 Control Plane Overview

LDP RSVP-TEBGP

ISIS

OSPF

EIGRPStatic

FIB Adjacency

LC NPU

ARP

LSD RIB

AIBSW FIB

LC CPU

RP

AIB: Adjacency Information Base

RIB: Routing Information Base

FIB: Forwarding Information Base

LSD: Label Switch Database

Over internal EOBC

Hierarchical FIB table

structure for prefix

independent

convergence: TE/FRR,

IP/FRR, BGP, Link

bundle

IOS-XR 2-Stage Lookup Packet FlowUnicast Packet Flow• Ingress lookup yields packet egress port and applies ingress features

• Egress lookup performs packet-rewrite and applies egress features

• All ingress packets are switched by Central switch fabric

Switch Fabric

Switch Fabric

3x

10G3x10GE

SFP +

3x10GE

SFP +

Typho

on

Typho

on

3x

10G

3x

10G3x10GE

SFP +

3x10GE

SFP +

Typho

on

Typho

on

3x

10G

3x

10G3x10GE

SFP +

3x10GE

SFP +

Typho

on

Typho

on

3x

10G

3x

10G3x10GE

SFP +

3x10GE

SFP +

Typho

on

Typho

on

3x

10G

FIA

FIA

FIA

FIA

Sw

itch

Fab

ric

AS

IC

Ingress

Typhoo

n

FIA

FIA

FIA

FIA

Sw

itch

Fa

bric

AS

ICEgress

Typhoo

n

Ingress

Typhoo

n

Egress

Typhoo

n

100GE

MAC/P

HY

100GE

MAC/P

HY

100

G

100

G

100

G

100

G

from wire

Ingress NPU

Egress NPU

TCAM rxIDB L3FIB

Packet classification

Source interface info

L3 FIB lookup

Next-hop

Packet rewriteSystem headers added

rewrite

L3FIB

L3 FIB lookup

Next-hop

Switch Fabric Port (egress NPU)

destination interface info

Fabric

ECH type: tell egress NPU type of lookup it should execute

L3 Unicast ForwardingPacket Flow (Simplified) Example

lookup key

L3: (VRF-ID, IP DA)

SFP

Rx LAG hashing

LAG SFPLAGID

ACL and QoS Lookup

also happen in parallel

ECH Type: L3_UNICAST

SFP

ECH Type: L3_UNICAST

=> L3FIB lookuprewrite txIDB

Tx LAG hashing

LAG

to wire

ACL and QoS Lookup

happens before rewrite

tx-adj

rx-adj

L3 Multicast Control Plane

1. Incoming joins (IGMP, PIM, MLD, MLDP, etc …)2. NPU punts joins to RP directly bypassing LC CPU3. Each protocol updates multicast info (VRF, route, olist, etc …) to MRIB4. MRIB downloads all multicast info (VRF, route, olist, etc …) to MFIB in each LC5. MFIB programs HW with multicast info: Typhoon NPU, FIA and LC Fabric

L2 Multicast Control Plane

1. Incoming joins (IGMP, PIM, MLD, etc …)2. NPU punts joins to RP directly bypassing LC CPU3. Each snooping protocol updates L2FIB (mrouter, port-list, input port, BD, etc … )4. RP L2FIB downloads L2 multicast info to each LC L2FIB5. LC L2FIB programs HW with multicast info: Typhoon NPU, FIA and LC Fabric

Multicast Replication Model Overview2-Stage Replication

Multicast Replication in ASR9k is like an SSM tree

2-stage replication model:

1. Fabric to LC replication

2. Egress NP OIF replication

ASR9k doesn’t use inferior “binary tree” or “root uniary tree” replication model

FGID (Slotmask)

© 2006 Cisco Systems, Inc. All rights reserved. Cisco Confidential EDCS:xxxx 5

LC 7

LC 6

LC 5

LC 4

RSP

0

RSP

1

LC 3

LC 2

LC 1

LC 0

Logical

Slot

9 8 7 6 5 4 3 2 1 0

Slot Slot Mask

Logical Physical Binary Hex

LC7 9 1000000000 0x0200

LC6 8 0100000000 0x0100

LC5 7 0010000000 0x0080

LC4 6 0001000000 0x0040

RSP0 5 0000100000 0x0020

RSP1 4 0000010000 0x0010

LC3 3 0000001000 0x0008

LC2 2 0000000100 0x0004

LC1 1 0000000010 0x0002

LC0 0 0000000001 0x0001

Target Linecards FGID Value (10 Slot Chassis)

LC6 0x0100

LC1 + LC5 0x0002 | 0x0080 = 0x0082

LC0 + LC3 + LC7 0x0001 | 0x0008 | 0x0200 = 0x0209

FGID Calculation Examples

FGIDs: 10 Slot Chassis

LC 3

LC 2

LC 1

LC 0

RSP 1

RSP 0

Logical

Slot

Phy

Slot

Number

5

4

3

2

1

0

Slot Slot Mask

Logical Physical Binary Hex

LC3 5 0000100000 0x0020

LC2 4 0000010000 0x0010

LC1 3 0000001000 0x0008

LC0 2 0000000100 0x0004

RSP1 1 0000000010 0x0002

RSP0 0 0000000001 0x0001

FGIDs: 6 Slot Chassis

ASR9k NG-MVPN (MLDP/P2MP-TE Packet FlowImposition PE

ASR9k NG-MVPN (MLDP/P2MP-TE Packet FlowDisposition PE

2 Stage Forwarding Model VPLS Packet Flow Example (Known Unicast)

Switch Fabric

Switch Fabric

3x

10G3x10GE

SFP +

3x10GE

SFP +

Typho

on

Typho

on

3x

10G

3x

10G3x10GE

SFP +

3x10GE

SFP +

Typho

on

Typho

on

3x

10G

3x

10G3x10GE

SFP +

3x10GE

SFP +

Typho

on

Typho

on

3x

10G

3x

10G3x10GE

SFP +

3x10GE

SFP +

Typho

on

Typho

on

3x

10G

FIA

FIA

FIA

FIA

Sw

itch

Fab

ric

AS

IC

Ingress

Typhoo

n

FIA

FIA

FIA

FIA

Sw

itch

Fa

bric

AS

ICEgress

Typhoo

n

Ingress

Typhoo

n

Egress

Typhoo

n

100GE

MAC/P

HY

100GE

MAC/P

HY

100

G

100

G

100

G

100

G

ETH ETH VC ETH VC Local Fab Hdr ETH VC Local Fab Hdr ETH VC LDPETH L2

Rewrite

IOS-XR Architecture1. 32-bit and 64-bit OS 2. IOS-XRv 9000 Virtual Forwarder3. Netconf/Yang Programmability

IOSd

Kernel

Linux-BinOSH

oste

d A

pp

1

Hoste

d A

pp

2

Operational Infra

Virtualization Layer

IOS “Blob”

Cisco IOS Cisco IOS-XE Cisco IOS-XR Cisco Virtual IOS-XR

XR Code v2

Kernel

Linux, 64bit

Distributed Infra

BG

P

OS

PF

PIM

AC

L

QoS

LP

TS

SN

MP

XM

L

NetF

low

Kernel

Linux, 64bit

System

Admin

XR Code v1

Kernel

Linux, 64bit

Distributed Infra

BG

P

OS

PF

PIM

AC

L

QoS

LP

TS

SN

MP

XM

L

NetF

low

Kernel

QNX, 32bit

Distributed Infra

BG

P

OS

PF

PIM

AC

L

QoS

LP

TS

SN

MP

XM

L

NetF

low

Control

Plane

Data

Plane

Mgmt

Plane

1990s 2000s 2003-14 Present Day

Cisco IOS – A Recap

Incremental Development, with Industry leading investment protection

32-bit XR Key Concepts

• Two Stage Commit

• Config History Database

• Rollback

• Atomic vs. Best Effort

• Multiple Config Sessions

• Etc, etc …

XR Command Modes

Exec – Normal operations - monitoring routing and CEF

Config – Configuration for L3 Node

Admin – Chassis operations, outside of SDRs

RP/0/RP0/CPU0:router#show ipv4 interfaces brief show running-configshow install active show cef summary location 0/5/CPU0

Admin Config

RP/0/RP0/CPU0:router(config)#router bgp 100 taskgroup admins policy-map foompls ldp ipv4 access-list block-junk

RP/0/RP0/CPU0:router(admin)#show controllers fabric plane all (CRS) config-register 0x0 show controllers fabric clock (12K) install add (also in SDR)

RP/0/RP0/CPU0:router(admin-config)#sdr backbone location 0/5/* pairing reflector location 0/3/* 0/4/*

Cisco IOS XR (Virtualized)

Virtualization Layer

Cisco Virtual IOS-XR

XR Code v2

Kernel Linux, 64bit

Distributed Infra

BG

P

OS

PF

PIM

AC

L

QoS

LP

TS

SN

MP

XM

L

NetF

low

Kernel Linux, 64bit

System

Admin

XR Code v1

Kernel Linux, 64bit

Distributed Infra

BG

P

OS

PF

PIM

AC

L

QoS

LP

TS

SN

MP

XM

L

NetF

low

Distributed system architectureScale, fault isolation, control plane expansion

Enhanced OS InfrastructureScale, SMP support, 64-bit CPU Support, Open Source apps

VirtualizationISSU, control and admin plane separation, simplifies SDR

Architecture

ISSU and HA ArchitectureProvide Zero Packet Loss (ZPL) and Zero Topology Loss (ZTL) to

avoid service outages

Fault ManagementCarrier class fault handling, correlation and alarm management

Available since IOS XR 5.0 on NCS6KOther platforms rollout – NCS4K, ASR9K, Skywarp, Fretta,

Sunstone and Rosco

Forwarding and Feature Path

• ACLs

• uRPF

• Marking, Policing

• IPv4, IPv6, MPLS

• Segment routing

• BFD

IOS-XRv 9000: Efficient Virtual ForwarderHigh-end Feature Rich Data Plane on x86

Innovative Virtual Forwarder

x86-optimized SW based hardware assists:

• 2x10G Line Rate FDX PCIe pass-through

• SW hierarchical traffic manager with 3 level HQoS, 512K queues @ 20G FDX performance per core

• SW policers that is color aware and nearly 4x faster than DPDK based SW routers

• SW TCAM with logical super key & heuristic cuts algorithms

Data plane optimized for fast convergence 20Gbps+ forwarder with features for IMIX traffic with features (on a single socket)

Portable 64bit C-code (to ARM based platforms)

Common code base with Cisco nPower X family

IOS-XRv Control Plane

RX &

Interface

Classification

Traffic

Manager

& TX

Forwarding

& Features

TCAM PLU

TM

Pkt. replication

IOS-XRv 9000

IOS-XRv Virtual Forwarder

Hierarchical QOS Scheduler

• 30Gbps+ throughput

on a single CPU core

• ½ million Queues

• 3-Layer H-QOS

High speed interface

classification

and fine grained load balancing

SW based HW Assists

NETCONF/YANG – Supported on XR

NETCONF – NETwork CONFiguration Protocol

• Network Management protocol – defines management operations

• Initial focus on configuration, but extended for monitoring operations

• First standard - RFC 4741 (December 2006)

• Latest rev is RFC 6241 (June 2011)

• Does not define content in management operations

YANG – Yet Another Next Generation

• Data modeling language to define NETCONF payload

• Defined in the context of NETCONF, but not tied to NETCONF

• Addresses gaps in SMIv2 (SNMP MIB language)

• Previous failed attempt – SMI NG

• First approved standard - RFC 6020 (October 2010)

N

E

T

C

O

N

F

YANG Data

Complete Your Online Session Evaluation

Don’t forget: Cisco Live sessions will be available for viewing on-demand after the event at CiscoLive.com/Online

• Give us your feedback to be entered into a Daily Survey Drawing. A daily winner will receive a $750 Amazon gift card.

• Complete your session surveys though the Cisco Live mobile app or your computer on Cisco Live Connect.

Thank you

cisco asr 9000 system architectured2zmdbbm9feqrf.cloudfront.net/2015/usa/pdf/brkarc-2003.pdf ·...

Documents