AT-8948 Switch for Service ProvidersArchitecture and Switching

Introduction

This presentation covers the following topics:• Switch architecture• Power supplies• Front-to-back cooling• Switching tables and traffic flow• Switching features• CompactFlash

Switch Architecture

48 x 10/100Base-T RJ45

Front to back cooling

4 x GbE SFP ports

Console port RJ45

Compact Flash slot

1RU

Switch Architecture

PHYs

IPv6 Accelerator Interface (10Gbps)

CPU – 400Mhz

DRAM – 128MB(up to 256 or 512MB)

Switch Chip

Marvel Prestera EX115 (37.6Gbps 13Mpps)

Dual Hot Swappable PSUs

PAC interface

32MB Packet Buffer Memory

32MB Flash

Switch Architecture

10/100 PHY

MarvellPacket Processor

Ports 1-16 Ports17-33 Ports 33-48

400MHzCPU

INTERFACE FOR IPv6 ACCELERATOR CARD

Route LookupMemory

PacketBuffer

PAC CardConnector

SDRAM (DIMM CONNECTOR)

EnviromentalMonitoring

CompactFlash

32 MByteFixed

FLASH

GigabitPHY

10/100 PHY

10/100 PHY

10/100 PHY

10/100 PHY

10/100 PHY

Hot-Swappable PSULoad Sharing

PSU 1

Hot-Swappable PSULoad Sharing

PSU 2

PO

RT49

(SFP)

PO

RT50

(SFP)

PO

RT51

(SFP)

PO

RT52

(SFP)

AC OR DC AC OR DC

BatteryBackedSRAM

Power Supply Units - Overview

Power supply design characteristics include: A choice of AC or DC input PSUs Redundancy, load sharing, and hot swappability Performance monitoring

There are three devices that may be fitted into the PSU bays: AC power supply DC power supply Fan only module (FOM)

It’s important that both PSU bays are occupied to assist with internal cooling. If only one PSU is fitted, a FOM must be fitted in the other bay.

Devices – what can be mounted in the PSU bays?

FOM

AC Power Supply Features

100-240v AC auto-ranging input voltage High efficiency (>80%) Class A EMC Power factor correction Current overload protection Protection from current spikes Over temperature alarm & shutdown Command-line access for testing PSU operation

DC Power Supply Features

48v DC input – compatible with -48v power distributiuon bus design High efficiency (>80%) Class A EMC Current overload protection Inrush current limiting Protection from current spikes Over temperature alarm & shutdown Command-line access for testing PSU operation

FOM – Fan Only Module

The FOM has two cooling fans and is powered via a back-feed from the PSU fitted in the other bay

Power Supply Redundancy, Load Sharing

and Hot-SwappabilityOne PSU is adequate to supply the switch’s power. If two are fitted they:

– will load share– can be hot-swapped

If a PSU’s ‘power good’ LED is lit, it is supplying current to the unit

What is the command to check load-sharing

performance?Release 2.6.2 includes the command: ACTIVATE SYSTEM PSU=[BAY1|BAY2|ALL] TEST=[LOADSHARE]

This allows the switch to check if each PSU can supply the full load

Ensures that the remaining PSU will cope with the extra demand when the other PSU is removed

Can you remove and replace PSUs while the

switch is operating?Yes, PSUs are hot-swappable

To meet regulatory requirements, two warnings are printed on the PSUs:– Caution: Disconnect power cord prior to removal of PSU– Caution: Disconnect all power cords to disable system power

Command-Line PSU Information

Some information is available via the command line interface:

SHOW SYSTEM

This command displays the module type, revision and serial number of each PSU or FOM that is installed, and the current condition of the modules

Example output from SHOW SYSTEM

command with PSU information highlighted

Front-to-Back Cooling

The AT-8948 uses a linear airflow arrangement, which is designed to reduce the amount of warm air being re-circulated

Air is drawn in from the front and expelled from the rear, due to the action of the PSU and FOM fans

The Benefits of Front-to-Back Cooling

No need for ventilation clearances above, below, left or right of the switch

Cabling clearances at front and rear of the switch are adequate for ventilation

Ambient operating temperature range is 0 to 50C

Switching Tables

The switch makes its forwarding decisions based on the entries in its switching tables

There are 18 hardware and 8 software tables Some statistics:

– 4K IP interfaces– 4K entries in the Multicast table– 16K entries in the MAC table– 256K entries in the IP route table

Trunking

Trunking allows a number of ports to be configured to join together to make a single logical connection of higher bandwidth

While the trunk group is logically a single connection, physically it involves up to four separate links

Trunking

Hashing of information in the L2, 3,and 4 packet headers divides traffic between the ports in the trunk group

PAYLOAD DATA

SCRMAC

DESMAC SCR

IPDESIP SRC

PortDESPort

When configuring trunking be aware that:

A maximum of 7 groups may be created

A trunk group may include a maximum of 4

ports

Ports in a trunk group need not be contiguous

Ports in a trunk group must belong to the same VLAN and have the same tagging status

When configuring trunking be aware that (continued) All ports in a trunk group must be added to

VLANs together, and can only be removed from a VLAN as a group

If the tagging status of the ports in a trunk group is changed, it must be changed for all ports in the trunk group at the same time

When configuring trunking be aware that (continued) Ports in a trunk group are set to autonegotiate

at the trunk speed at full duplex

When a port is added to a trunk group, the speed setting for the group overrides the speed setting previously configured for the port

When configuring trunking be aware that (continued) When a port is removed from a trunk group, the

port returns to its previously configured speed and duplex mode settings

A trunk group may not include both 10/100 Ethernet ports and Gigabit Ethernet ports

A trunk group may not include a mirror port

The port trunking algorithm used on the AT-8948 may be compatible with that used on third-party devices

When configuring trunking be aware that (continued)

VLAN Double Tagging Available in 2.6.2

An enterprise with multiple VLANs across multiple sites in a metropolitan area can use a public MAN to carry VLAN-tagged traffic between its sites

The public MAN carries traffic for multiple customers, so each frame must carry ‘customer-ID’ information

A second VLAN tag is inserted into each frame as it enters the public network and is removed on egress

VLAN Double Tagging

AT-8948

MetroEthernet network

AT-8948

AT-8948

Customer ASite 3

VLANs 1,2,3

Customer ASite 2

VLANs 1,2,3

Customer BSite 2

VLANs 1,2,3

Customer BSite 1

VLANs 1,2,3

Customer B Site 3

VLANs 1,2,3

Customer ASite 1

VLANs 1,2,3

SwiA_DoubleTag

VLAN Double Tagging

‘Familiar’ VLAN tagging: a 4-byte field is inserted into Ethernet frames

FCSPayload DataLEN / ETypeSADA

Payload DataLEN / ETypeSADA

6

66

4

4

6 2 46 to 1500

24 46 to 1500

OriginalEthernetFrame

VLAN-taggedEthernet Frame

Max Frame sizeis 1518 bytes

Max Frame sizebecomes 1522bytes

TAG FCS

SwiA_Frame1

VLAN Double Tagging

When a frame enters the public MAN, a second, ‘per-customer’ VLAN tag is inserted

SwiA-Frame2

Double taggedFrame withinprovider network

Max Frame sizebecomes 1526bytes

FCSPayload DataLEN / ETypeSADA TAGTAG

6 6 444 2 46 to 1500

Configuring VLAN double-tagging

Create the nested VLAN, specifying the customer-ID

CREATE VLAN={vlan-name VID=2..4094} NESTED

Configuring VLAN double-tagging

Two types of ports are defined: Customer ports connected to customers’ LANs:

ADD VLAN={vlan-name|2..4094} PORT=port-list NESTEDTYPE=CUSTOMER

Core ports connected to the public MAN ADD VLAN={vlan-name|2..4094} PORT=port-list

NESTEDTYPE=CORE

Configuring VLAN double-tagging

To operate with other vendors’ equipment, it may be necessary to change the ‘Ethertype’ value in the customer-ID tag from its default value of 0x8100:

SET SWI NESTEDTPID=TagNumber

VLAN Double Tagging

Customer-ID tag

ETypeTAG ControlIDENTIFIER

8 567 4 3 2 1 8 7 6 5 4 3 2 1

User Priority CFI 12bit VID (VLAN ID)

2 bytes 2 bytes

FCSPayload DataLEN / ETypeSADA TAGTAG

6 6 444 2 46 to 1500

Archswi_NTA3

Private VLANs Available in 2.6.2

A Private VLAN is a VLAN which contains a specified group of ports that are prevented from communicating with each other at Layer 2.

(Also known as a protected or port-protected VLAN)

Private VLANs

A typical application is in hotel installations where each room is serviced by one 10/100 Ethernet port, through which the hotel guest is able to access the Internet. In this situation it is undesirable to allow communication between rooms.

Private VLANs

One customer is not able to snoop on the traffic from any other, yet each customer is able to access another network (usually the Internet).

SwiA_PVlan1

INTERNET

X

Private VLAN 10110.1.1.99/24 10.1.1.100/24

AT-8948

AT-8948

Private VLANs

Ports that are members of a Private VLAN have one of two states, either 'private' (protected) or 'uplink'

Private ports cannot talk to other private ports but can talk to uplink ports. Uplink ports can talk to both private and other uplink ports, if they exist

Private VLANs

All traffic received on any private port in a Private VLAN is sent to the predefined uplink port, and only that uplink port, regardless of VLAN ID or MAC Destination address

Layer 2 traffic between private ports that are members of a Private VLAN is blocked

Configuring Private VLANs

Create a private VLAN using the command: CREATE VLAN=vlan-name VID=2..4094 PRIVATE

Add the private ports to the VLAN: ADD VLAN={vlan-name VID=2..4094} PORT=port-list

Add the uplink port(s) to the VLAN: ADD VLAN={vlan-name VID=2..4094} PORT=port-list UPLINK

If the uplink ‘port’ is a trunk group, the trunk group must be created before the ports are added to the private VLAN

CompactFlash

CompactFlash

A small removable mass storage device that uses FLASH memory

Memory that doesn’t require power from a battery to retain stored data

Used to expand the amount of FLASH memory available to store files on the switch

CompactFlash is:

CompactFlash

Two CompactFlash cards have been approved for the AT-8948:

AT-CF032A-n 32MB CompactFlash cardAT-CF0128A-n 128MB CompactFlash card

Where n is the number of cards in a package

(The size of the release file 89-261.rez is in the order of 6MB)

QoS

Feature– Eight priority/egress queues per port

Benefits– Allows traffic to be processed with up to eight levels of priority– Gives greater control to the network administrator– Increases differentiation of critical and non-critical network applications– Ensures availability of business-critical applications and services

Catalyst 3550/3750 Metro– 4 egress queues per port

QoS

Features– Bandwidth limiting down to 3Kbps, with burst limits – Bandwidth limit resolution down to 1Kbps

Benefits– Very precise control of bandwidth guarantees– Burst limits improve bandwidth limiting of TCP sessions (avoids bandwidth

flapping)

Catalyst 3750 Metro– rate limiting at 8 Kbps increments “The bandwidth limiting

capabilities of the 8948 are frightening!” – Senior

Test Engineer, 8948 Development Team, ATR.

QoS

Feature– 2 rates & 3 colours of bandwidth conformance

Benefits– Allows SPs to offer differentiated services based on SLA– Customers exceeding their guaranteed SLA bandwidth can be given

lower priority using re-marking– Non-conforming traffic can be identified through the entire network

Catalyst 3750 Metro– 2 rate 3 colour (CIR/EIR) rate limiting

QoS – Rate Metering

Bandwidth Class 3

Bandwidth Class 2

Bandwidth Class 1

Immediate Discard on Ingress

RED Discard

Re-mark

Re-mark

MAXBURSTSIZE

Re-markCommitted Information Rate (CIR)

Traffic Class Bandwidth

Time

MAXBANDWIDTH

MINBANDWIDTH

MINBURSTSIZE Committed Burst Size (CBS)

Excess Information Rate (EIR)

Excess Burst Size (EBS)

Re-marking options :

bandwidth class

DSCP

Egress Queue

VLAN Tag User Priority

QoS – Common CoS-based SLA

Service Class

Service Characteristics

802.1pID

Bandwidth Profile

Service Performa

nce

Premium Real-time IP telephony or IP video applications 6, 7 CIR

No EIR

Delay < 5msJitter < 1ms

Loss < 0.01%

SilverBursty mission critical data applications requiring low loss and delay (e.g., Storage)

4, 5 CIREIR

Delay < 5msJitter = N/S

Loss < 0.01%

BronzeBursty data applications requiring bandwidth assurances

3, 4 CIREIR

Delay < 15msJitter = N/S

Loss < 0.1%

Standard Best effort service 0, 1, 2 No CIRDelay < 30ms

Jitter = N/SLoss < 0.5%

IPv6 – Why?

Increased address space– IPv4: 32 bit address gives 4 billion addresses– IPv6: 128 bit address gives 340 billion billion billion billion

addresses!!!!

True end-to-end networking – Removes need for NAT

– Some countries in Asia, with only a small IPv4 address allocation, have NAT up to 6 layers deep!

– Some countries are running out of addresses now!

Increased securityBetter QoS – Flow labelsAutomatic configuration

IPv6

Feature– Dual IPv4 and IPv6 stack

Benefits– IPv6 routing in software as part of feature set – DHCPv6, RIPng,

multicasting– Ideal for early adopters looking for IPv6 compatibility now

Universities Government/Military

– Allows IPv4 and IPv6 to coexist in the same networks, easing transition to IPv6

IPv6

Feature– IPv6 accelerator card ACC-01

Benefits– Ideal for customers who demand IPv6

compatibility and high performance IPv6 routing– Provides wire speed hardware routing for IPv6

packets– Provides wire speed QoS – prioritisation and rate

limiting– Accelerates tunnelling– Can sell as upgrade option for AT-8948A

customers wanting future IPv6 capability– Future proofed

Start with IPv4 box Sell IPv6 upgrade later

IPv6

Feature– Tunnelling

Benefits– Future-proofs your network– Allows early adopters to implement IPv6 at the edge, retaining IPv4

core– IPv4 and IPv6 will co-exist for some time. Tunnelling eases

transition to IPv6.

IPv6 Tunnelling Application

IPv4 Internet

IPv4IPv4

IPv6IPv6

AT-8948AT-8948

End of presentation