IEEE2finalrtml ver2

8/7/2019 IEEE2finalrtml ver2

http://slidepdf.com/reader/full/ieee2finalrtml-ver2 1/25

A Distributed & ScalableRouting Table Manager

for the Next Generation of IP Routers

Sheetal V A

1BM09SCS10

I Sem M.Tech., CSE



Abstract

This article, describesThis article, describes

Benefits and limitations of a distributed router Benefits and limitations of a distributed router

design.design. Propose a distributed architecture for the RTM.Propose a distributed architecture for the RTM.

Presents a comparative scalability evaluation of thePresents a comparative scalability evaluation of theproposed distributed architecture with a centralizedproposed distributed architecture with a centralizedone, in terms of required memory and computingone, in terms of required memory and computing

resources.resources. costcost--effective approach, where legacy routers iseffective approach, where legacy routers is

replaced by a nextreplaced by a next--generation router with a verygeneration router with a verylarge switching capacity.large switching capacity.



First and Second generations of routersFirst and Second generations of routers

Single central processor with multiple line cardsSingle central processor with multiple line cards

Shared bus interconnectionShared bus interconnection Performance dependent on throughput of the busPerformance dependent on throughput of the bus

and processor speed.and processor speed.

Does not meet high bandwidth requirements.Does not meet high bandwidth requirements.

Third / Current generation routersThird / Current generation routers Uses switch fabric connection between line cardsUses switch fabric connection between line cards

Forwarding engine on each line cardForwarding engine on each line card

Legacy routers



Next generation routers

Switch based architecturesSwitch based architectures

3 types of cards3 types of cards

Line card with ingress and egress networkLine card with ingress and egress network

Processor that perform packet forwarding,Processor that perform packet forwarding,

classification, policing and scheduling.classification, policing and scheduling.

Supports multiple gigabit interfaces.Supports multiple gigabit interfaces.

Ingress NP contains FIT to determine packetIngress NP contains FIT to determine packet

destination and traffic manager for policing anddestination and traffic manager for policing and

buffer managementbuffer management

Egress NP performs queuing and schedulingEgress NP performs queuing and scheduling



Contd..

Control card runs routing protocols (likeControl card runs routing protocols (like

BGP, OSPF, MPLS), RTM and CLIBGP, OSPF, MPLS), RTM and CLI

CPU has higher processing power CPU has higher processing power Provides traffic management for control trafficProvides traffic management for control traffic

from control card to line cardfrom control card to line card

The Control and Line cards areThe Control and Line cards are

interconnected by a scalable switch fabricinterconnected by a scalable switch fabric



Architecture of thenext generation router



Routing Table Manager

Route table manager Route table manager Main software component of the router Main software component of the router

Builds forwarding table from the routing databaseBuilds forwarding table from the routing database

which is populated by routing protocolswhich is populated by routing protocols Routes added to FIT can be selected based onRoutes added to FIT can be selected based on

preference associated with each routing protocolpreference associated with each routing protocol

Update the FIT based on change in routingUpdate the FIT based on change in routingdatabasedatabase

Notify routing protocols regarding physical linkNotify routing protocols regarding physical linkstatus, bandwidth change.status, bandwidth change.

FIT is downloaded to all line cardsFIT is downloaded to all line cards



RTM in a non-distributedrouting architecture



Current RTM architectures

Non distributed architectureNon distributed architecture

RTM only on control card.RTM only on control card.

Best routes from different routing protocolsBest routes from different routing protocolsare selected based on protocol preferenceare selected based on protocol preferenceand populated in FIT.and populated in FIT.

Resiliency requires additional control cardsResiliency requires additional control cards

Not scalable for large no of routesNot scalable for large no of routes



Distributed RTM Architecture



RTM distributed on the control card

RTM for each protocol module denoted by IGPRTM for each protocol module denoted by IGP/ EGP./ EGP.

One instance of GOne instance of G--RTM that collects bestRTM that collects bestroutes from protocol RTM and builds FIT.routes from protocol RTM and builds FIT.

Protocol RTM & FIT on control card areProtocol RTM & FIT on control card areupdated to line cards.updated to line cards.

Resiliency is improved because routingResiliency is improved because routingprotocols can use IGP / EGP RTMs when Gprotocols can use IGP / EGP RTMs when G--RTM fails.RTM fails.



Contd

Issues:Issues:

IGP/EGP are independent process runningIGP/EGP are independent process running

on same control card results in higher on same control card results in higher resource utilizationresource utilization

Overloading on Control card as the number Overloading on Control card as the number

of routes increasesof routes increases

If the protocols are distributed on LC thenIf the protocols are distributed on LC then

RTM modules need to be migrated to LC¶sRTM modules need to be migrated to LC¶s



Proposed RTM distributed architecture

Some of the RTM functionality is moved from CC to LCSome of the RTM functionality is moved from CC to LC

Distributed architecture consists of two mainDistributed architecture consists of two maincomponentscomponents LCLC--RTMRTM

Each LC handles LCEach LC handles LC--RTM process.RTM process.

LCLC--RTM obtains route information from protocols running on LCRTM obtains route information from protocols running on LCand computes best routesand computes best routes

GG--RTMRTM

Runs on control card and obtains information from LCRuns on control card and obtains information from LC--RTM toRTM toupdate routing table and FITupdate routing table and FIT

Manages configuration of static routes & traffic engineering (TE)Manages configuration of static routes & traffic engineering (TE)based Routesbased Routes



Proposed Distributed RTM architectureon the control card and line card



Distribution of RTM.



Architecture of LC-RTM



Architecture of G-RTM



Proposed G-RTM & LC-RTM

Inter card communication between LCInter card communication between LC--RTMsRTMs

on LC or between Gon LC or between G--RTM and LCRTM and LC--RTMRTM

achieved by a distribution services (DS)achieved by a distribution services (DS) DS provides sync mechanism for moduleDS provides sync mechanism for module

activation, monitoring and state transitionsactivation, monitoring and state transitions

DS maintains distribution database thatDS maintains distribution database that

enables modules to get required dataenables modules to get required data



Distributed Model is basedon following aspects

Link State NotificationsLink State Notifications

RTM are notified of the changes in routingRTM are notified of the changes in routing

information for computing best routesinformation for computing best routes

AdvertisementsAdvertisements

When LC detects change in link state thenWhen LC detects change in link state then

RTM broadcasts a notification and routingRTM broadcasts a notification and routingtables are recalculated and neighbor routerstables are recalculated and neighbor routers

are notified.are notified.



Contd

Path ComputationsPath Computations

Routing protocol modules receive updateRouting protocol modules receive update

information regarding link modificationsinformation regarding link modifications

LCLC--RTM running on the Line card is notifiedRTM running on the Line card is notified

Notification is send to GNotification is send to G--RTM or Master LineRTM or Master Line

card based on protocol identificationcard based on protocol identification New route or updated route is registered toNew route or updated route is registered to

the FIT located on line cards through Gthe FIT located on line cards through G--RTMRTM



Advantage of Proposed Architecture

ScalabilityScalability Path computation processing shared by CC and LCPath computation processing shared by CC and LC

RTM functions are distributed to allow CC to perform other RTM functions are distributed to allow CC to perform other

tasks like router management and user interactiontasks like router management and user interaction High AvailabilityHigh Availability

Router information and databases have back up on LCs thereRouter information and databases have back up on LCs thereby providing redundancyby providing redundancy

Control card errors do not affect packet forwarding on LCsControl card errors do not affect packet forwarding on LCs

RobustnessRobustness The path computation is performed on set of LCs instead of The path computation is performed on set of LCs instead of

whole routers which leads to faster topology convergencewhole routers which leads to faster topology convergence

LCLC--RTM reacts rapidly to the physical link modificationRTM reacts rapidly to the physical link modification



Scalability Evaluation

Distributed ArchitectureDistributed Architecture

reduces the traffic on the switch fabricreduces the traffic on the switch fabric

Memory requirement on the control card isMemory requirement on the control card isreduced considerablyreduced considerably

CPU utilization on control card is much lower CPU utilization on control card is much lower compared to centralized architecturecompared to centralized architecture

Allows the load on cc to be transferred toAllows the load on cc to be transferred tomaster line card so control card congestionsmaster line card so control card congestionsis avoidedis avoided



Conclusion

The RTM is one of the most important

components of a router.

This article, presents distributed architecturemodel for the RTM for next-generation IP

routers.

Routes can be computed more efficiently and

in a scalable manner, based on interfacesbetween the LC-RTMs and routing protocols

running on line cards.



References

[1]A. Zini, Cisco IP Routing , Addison-Wesley, 2002, pp. 80±111.

[2] O. Hagsand, M. Hidell, and P. Sjodin, ³Design and Implementation of aDistributed

Router,´ Proc. 5th IEEE Int¶l. Symp. Signal Processing and Info.

Tech., Dec. 2005, pp. 227±32.

[3] A. Csaszar et al., ³Converging the Evolution of Router Architectures and IP

Networks,´ IEEE Network Mag., vol. 21, no. 4, July±Aug. 2007.

[4] H. J. Chao and B. Liu, High Performance Switches and Routers, Wiley-Interscience, 2007.

[5] Cisco Systems, ³Cisco 12000 Series Internet Router Architecture´;

http://www.cisco.com

[6] H. Kaplan, ³Non-Stop Routing Technology,´ white paper, Avici Systems Inc.,

2002.

[7] M. Leelanivas, Y. Rekhter, and R. Aggarwal, ³Graceful Restart Mechanism

for Label Distribution Protocol,´ IETF RFC 3478, Feb. 2003.[8] M. Deval et al., ³Distributed Control Plane Architecture for Network Elements,´

Intel Tech. J., vol. 7, no. 4, 2003.

[9] K. K., Nguyen et al., ³Towards a Distributed Control Plane Architecture for

Next Generation Routers,´ ECUMN 2007 , France, Feb. 2007.



THANK YOU!

IEEE2finalrtml ver2

Documents