An Approach For Intelligent Traffic Splitting For Sudden Changes Of Traffic Dynamics

DOI : 10.17577/IJERTV2IS60171

Download Full-Text PDF Cite this Publication

Text Only Version

An Approach For Intelligent Traffic Splitting For Sudden Changes Of Traffic Dynamics

GanaSindu. K. S1,SmithaShekar. B 2,Harish. G 2

1Department Of Computer Science, IV Sem, M.Tech Dr. Ambedkar Institute of Technology

Bangalore, Karnataka, India.

2Department Of Computer Science, Asst. Professor Dr. Ambedkar Institute of Technology

Bangalore, Karnataka, India.

AbstractIn the network management process, dynamically changing traffic is a very difficult task to manage and it leads to many problems which will decrease the network performance and increase delay. Mechanisms are needed that can handle traffic load dynamics in scenarios with sudden changes in traffic demand and dynamically distribute traffic to benefit from available resources.

In the previous work [1], AMPLE (Adaptive Multi- toPoLogy traffic Engineering) is introduced which consists of two distinct phases to achieve TE objectives. First is offline network dimensioning through link weight optimization for achieving maximum intra-domain path diversity across multiple routing topologies and second is adaptive traffic splitting ratio adjustment across these routing topologies for achieving dynamic load balancing in case of unexpected traffic dynamics.

In this paper another module which is traffic analyzer, which help in managing the traffic and study its nature and flow, and it reduces the burden from adaptive traffic controller. The traffic analyzer will help in sending the traffic in same order as it was received, and study its behavior so that it is easy to predict to some extent the possible future traffic flow. This will increase the performance of adaptive traffic controller and help to manage unexpected traffic flow.

KeywordsMulti Topology,dynamic traffic, routing Topology, load balancing, Traffic analyzer.


    Research in the Traffic Engineering (TE) field has beencarried out for years. Solutions exist, but few ofthese are actually used by operators to manage their network and one

    reason is that these methods are implementedfor research and simulation purposes. It is considered difficultto integrate these

    methods in an operational environment.Intra-domain Traffic Engineering (TE) based on IGPs such as OSPF and IS-IS hasrecently been receiving numerous attentions in the Internet research community.In order to achieve near-optimal or even optimal network performance, it is suggestedthat both IGP link weights and traffic splitting ratio need to be optimizedsimultaneously based on the Traffic Matrix (TM) [11],and the network topology asinput. However, this is only applicable to offline TE where knowledge of theestimated TM is assumed a priori. Unfortunately, this assumption is usually not validin real operational networks given frequent presence of traffic dynamics such asunexpected traffic spikes that are difficult to anticipate.

    As a result, the absence ofaccurate traffic matrix estimation may lead the offline TE approaches to performpoorly. The most straightforward approach for handling this is to reassign IGP linkweights dynamically in reaction to the monitored dynamics. However, re-assigninglink weights on the fly may cause transient forwarding loops during the convergencephase, which often leads to service disruptions and traffic instability.

    In previous work [1], AMPLE (Adaptive Multi-toPoLogy traffic Engineering),a novel IGP TE approach that is capable of adaptively handling traffic dynamics in operational IP networks. Instead of re-assigning IGP link weights in response to traffic fluctuations, we adopt multi-topology IGPs (MT- IGPs) such as MT-OSPF and M-ISIS as the underlying routing platform to enable path diversity, based onwhich adaptive traffic splitting across multiple routing topologies is performed fordynamic load balancing. AMPLE consists of two distinct phases to achieve our TEobjectives. First, the offline phase (e.g., at a weekly or monthly timescale) focuses onthe static dimensioning of the underlying network, with MT-IGP link weightscomputed for maximizing intra-domain path diversity across multiple routingtopologies. Since the objective is to

    obtain diverse IGP paths between eachsource/destination pair, the computation of MT-IGP link weights is actually agnostic to any traffic matrix. Once the optimized link weights have been deployed in thenetwork, an adaptive TE performs traffic splitting ratio adjustment for loadbalancing across diverse IGP paths in multiple routing topologies, according to the

    up-to-date monitored traffic conditions.

    This adaptive TE aims to efficiently handletraffic dynamics at short time-scale such as hourly or even in minutes. Given the factthat traffic dynamics are common in operational IP networks, our proposed approachprovides a promising and practical solution that allows network operators toefficiently cope with these dynamics that normally cannot be anticipated in advance.The contributions of our work can be summarized as follows. First of all, AMPLEdoes not require frequent and on-demand re-assignment of IGP link weights, thusminimizing the undesired transient loops and traffic instability. Second, theoptimization of the MT-IGP link weights does not rely on the availability of trafficmatrix a priori, which plagues existing IGP TE solutions due to inaccuracy of trafficmatrix estimations. Third is that another module called Traffic Analyzer is used to improve the efficiency and manage the traffic flow. Finally, our experiments based on real network topologies andtraffic matrices have shown that AMPLE has a very high chance of achieving nearoptimalperformance with only a small number of routing topologies.


    For a network operator it is important to analyse and tunethe performance of the network in order to make the best useof it [12]. The process of performance evaluation and optimizationof operational IP-networks is often referred to as trafficengineering. One of the major objectives is to avoid congestionby controlling and optimizing the routing function.The traffic engineering process can be divided in three partsas illustrated in Figure 1.

    Figure 1.

    The first step is the collectionof necessary information about network state. To be specific,the current traffic situation and network topology. Thesecond step is the optimisation calculations. And finally,the third step is the mapping from optimization to routingparameters. Current routing protocols are designed to besimple and robust rather than to optimize the resource usage.

    The two most common intra-domain routing protocolstoday are OSPF (Open Shortest Path First) and IS-IS (IntermediateSystem to Intermediate System). They are bothlink-state protocols and the routing decisions are typicallybased on link costs and a shortest (least-cost) path calculation.While this approach is simple, highly distributed andscalable these protocols do not consider network utilizationand do not always make good use of network resources. Thetraffic is routed on the shortest path through the networkeven if the shortest path is overloaded and there exist alternativepaths. With an extension to the routing protocols likeequal-cost multi-path (ECMP) the traffic can be distributedover several paths but the basic problems remain. An underutilizedlonger path cannot be used and every equal costpath will have an equal share of load.


A classification of traffic engineering [12], schemes is possiblealong numerous axis. Our framework is intended to facilitatethe analysis and help us identify the requiements for traffic engineering.

  1. Optimize Legacy routing vs. Novel routing mechanisms. One approach is to optimize legacy routing protocols. Theadvantage is easy deployment of the traffic engineeringmechanism. However, the disadvantage is the constraintsimposed by legacy routing.

  2. Centralizedvs. Distributed solutions.

    A centralized solutionis often simpler and less complex than a distributed,but is more vulnerable than a distributed solution.

  3. Local vs. Global information.

    Global information ofthe current traffic situation enables the traffic engineeringmechanism to find a global optimum for the load balancing.The downside is the signalling required to collect theinformation. In addition, in a dynamic environment, the informationquickly becomes obsolete.

  4. Off-line vs. On-line traffic engineering.

    Off-line trafficengineering is intended to support the operator in the managementand planning of the network. On-line traffic engineeringon the other hand, reacts to a signal from the networkand perform some action to remedy the problem.


      As already mentioned, AMPLE encompasses four distinct tasks, namely (1)offline network dimensioning through link weight optimization for achievingmaximum intra-domain path diversity across multiple MT-IGP routing topologies, and (2) adaptive traffic splitting ratio adjustment across these routing topologies forachieving dynamic load balancing in case of unexpected traffic dynamics.

      1. Network monitoring is used to get the volume of the traffic flow in the network and it constantly monitor the network and the amount of the traffic flow.

      2. Another module is proposed which is the Traffic Analyzer which is used to manage the traffic flow and maintain the correct order, which is shown in figure 2.

      Figure 2.


In Offline link weight optimization, the binary metricof Full Degree of Involvement (FDoI) [1] [2] , to evaluatethe overall path diversity for a given MT-IGP linkweight configuration. More specifically, the FDoIvalue for a link with respect to an S-Dpair is setto 1 if this link is shared by the shortest IGPpaths across all VRTs for that S-D pair, otherwiseit is set to

  1. Thefundamental idea behind this scheme follows thestrategy of offline provisioning of multiple diversepaths in the routing plane and online spreadingof the traffic load for dynamic load balancing inthe forwarding plane. Theapproach can be briefly described as follows.MT-IGPs are used as the underlying routingprotocol for providing traffic-agnostic intradomainpath diversity between all source-destinationpairs. With MT-IGP routing, customertraffic assigned to different virtual routingtopologies (VRTs) follows distinct IGP pathsaccording to the dedicated IGP link weight configurationswithin each VRT.

    Our ultimate objective is to minimize the chance that a single link is shared by allrouting topologies between each source- destination pair. The objective is to avoid introducing critical links with potential congestion where the associated source destination pairs cannot avoid using it no matter which routing topology is used. The Full Degree of Involvement (FDoI), which indicates whether a critical link l is included in the IGP paths between source-destination pair

    (u, v) in all routing topologies:

    , = = ||

    , = = ||

    The optimization objectiveof OLWO is to minimize the sum of FDoIvalues across all network links with regard to allS-D pairs. If this sum is equal to 0, then no criticallink is formed given the underlying MT-IGPlink weights, which means that at least one sourcein the network will always be able to find alternativepath(s) to bypass the over-loaded link givenany single link congestion scenario.


      Network monitoring [1], is responsible for collecting up-to- date traffic conditions in real-time andplays an important role for supporting the ATCoperations and it forms input for the traffic analyzer. AMPLE adopts a hop-by-hop based monitoring mechanism. The basic idea is that a dedicatedmonitoring agent deployed at every PoP node isresponsible for monitoring:

      • The volume of the traffic originated by thelocal customers toward other PoPs (intra-PoP traffic is ignored).

      • The utilization of the directly attachedinter-PoP links.


      Traffic monitoring and analysis is essential in order to moreeffectively troubleshoot and resolve issues when they occur, so as to not bring network services to a stand still for extended periods of time. Numerous tools are available to help administrators with the monitoring and analysis of network traffic.

      Network traffic analysis which provide a clear overview of the structure of traffic and enable the efficient detection of potential problems and irregularities. Network traffic analysis is the process of capturing network traffic and inspecting it closely to determine what is happening on thenetwork. A number of technologies have been developed to increase our understanding of the behavior of network traffic. It enables an overview of the statistics of the traffic passing through our network and is recommended for environments were the network devices can support this technology.

      1. Dos attacks (a large amount of traffic is being generated towards dns or email servers);

      2. Bandwidth abuse (such as, YouTube, face book, or torrent);

      3. Access to forbidden websites;

      4. Attempts to attack/access protected network devices; Advantages:

      1. Centralized data collection.

      2. The existing equipment may be used.

      3. Easy configuration.

      4. The possibility of collecting other parameters during communication, such as, delays, variation of delays

      or lost packets.

      The total action is given below figure 3.

      In the previous module only the volume and the amount of the traffic flow can be known and if this raw data is given to the Adaptive Traffic Controller means it will reduce the efficiency of ATC. So it is very necessary for the traffic analyzer to view all the traffic and manage it and then give that input to the ATC.

      It consists of two parts to manage the traffic

  2. Traffic Analyzer Manager- which do the function ofcollecting the incoming traffic and maintain the proper order of the incoming trafficso that whichever traffic link comes first will be the first one to be sent to the ATC for efficient traffic splitting.

  3. Traffic Analyzer Database- which is the central storage of the incoming traffic at that interval of time, and this database are mainly used for the prediction of the next incoming traffic and so can provide efficient steps for it.





The results of this analysis provide us with the following information:

1. Information on the total amount of traffic between individual subnets (bytes, packets, connections);

Figure 3.


    1. Information on the total amount of local traffic (protocols, servers and hosts);

    2. Information about external access to our network (protocol, service, host);

    3. A prediction of future traffic behavior;

    4. The existence of a virus in the network (a large amount of incoming or outgoing traffic is being

    5. Generated);

In this section [1] [2],adaptive adjustment of traffic splitting ratio at individual PoP source nodes. In a periodic fashion at a relatively short-time interval (e.g., hourly), by getting input from the Traffic Analyer, the central TE manager needs to perform the following three operations:

  1. Measure the incoming traffic volume and the network load for the currentinterval.

  2. Compute new traffic splitting ratios for all PoP nodes based on the measured traffic demand and the network load for dynamic load balancing.

  3. Instruct individual PoP nodes to enforce the new traffic splitting ratio over their locally originated traffic.

Given the optimized MT-IGP link weights producedby OLWO, Adaptive Traffic Control (ATC)can be invoked at short-time intervals duringoperation in order to re-optimize the utilizationof network resources in reaction to trafficdynamics.

The optimization objective of ATC isto minimize the Maximum Link Utilization(MLU), which is defined as the highest utilizationamong all the links in the network. Therationale behind ATC is to perform periodic andincremental traffic splitting ratio re-adjustmentsacross VRTs based on traffic pattern continuityat short a timescale, but without necessarilyperforming a global routing re- optimization processfrom scratch every time. To fulfill the second task, a Traffic EngineeringInformation base (TIB) is needed by the TE managerto maintain necessary network state basedon which new traffic splitting ratios are computed.

The structure TIB [1], which consists of two inter-related repositories,namely the Link List (LL) and S-D PairList (SDPL). The LL maintains a list of entriesfor individual network links. Each LL entryrecords the latest monitored utilization of a linkand the involvement of this link in the IGP pathsbetween associated S-D pairs in individual VRTs.More specifically, for each VRT, if the IGP pathbetween an S-D pair includes this link, then theID of this S-D pair is recorded in the LL entry. On the otherhand, the SDPL consists of a list of entries, eachfor a specific S-D pair with the most recently measured traffic volume from S to D.

Each SDPLentry also maintains a list of subentries for differentVRTs, with each recording the splitting ratioof the traffic from S to D, as well as the ID of thebottleneck link along the IGP path for that S-D

pair in the corresponding topology.

During each ATC interval, the TIB is updatedupon the occurrence of two events. First,upon receiving the link utilization report fromthe network monitoring component, the TEmanager updates the link utilization entry in theLL and the ID of the bottleneck link for eachS-D pair under each VRT in SDPL. Second,when the adaptive traffic control phase is completedand the new traffic splitting ratios arecomputed, the splitting ratio field in SDPL isupdated accordingly for each S- D pair undereach VRT.


In this paper AMPLE, a novel TE approach that enables dynamic load balancing in operational IP networks. Instead of frequently changing IGP link weights, we use multi-topology IGP routing protocols that allow adaptively splitting traffic across multiple routing topologies. Offline link weight optimization is performed in order to enable path diversity, followed by the adaptive control of traffic splitting across individual routing topologies according to the monitored traffic and also along with the network monitoring another module is used which is Traffic Analyzer for Traffic Management.


[1]Ning Wang, University of Surrey, Kin Hon Ho, Hong Kong Polytechnic University George Pavlou, University College London , "AMPLE: An Adaptive Traffic Engineering

System Based onVirtual Routing Topologies", IEEE 2012 Transactions on Communications, Volume: 55, Issue: 3 [2]N.Wang, K-H. Ho, and G. Pavlou, Adaptive MultitopologyIGP Based Traffic Engineering with Near- OptimalPerformance, Proc. IFIP Networking 2008.

[3] S. Uhliget al., Providing Public Intradomain TrafficMatrices to the Research Community, ACM SigcommComp. Commun. Rev. (CCR), vol. 36, no. 1, Jan. 2006,pp. 8386.

  1. B. Fortz and M. Thorup, Optimizing OSPF/IS-IS Weights in a Changing World, IEEE JSAC, vol. 20, no. 4, May 2002, pp. 75667.

  2. D. Xu, M. Chiang, and J. Rexford, Link-State Routing With Hop-By-Hop Forwarding Can Achieve OptimalTraffic Engineering, Proc. IEEE INFOCOM, Apr. 2008.

  3. M. Caesar et al., Dynamic Route Computation ConsideredHarmful, ACM Comp. Commun. Rev. (CCR), vol.40, no. 2, Apr. 2010, pp. 6671.

  4. N. M. MosharafKabirChowdury and R. Boutaba, ASurvey of Network Virtualization, Computer Networks,vol. 54, issue 5, Apr. 2010, pp. 86276.

  5. P. Psenaket al., Multi-Topology (MT) Routing inOSPF, RFC 4915, June 2007.

  6. A. Asgariet al., Scalable Monitoring Support forResource Management and Service Assurance, IEEENetwork Mag., vol. 18, issue 6, Nov. 2004, pp. 618.

  7. A. Kvalbeinet al., Multiple Routing Configurations for Fast IP Network Recovery, IEEE/ACM Trans. Net., vol.17, no. 2, 2009, pp. 47386.

  8. S. Balonet al., Traffic Engineering and Operational Network with the TOTEM Toolbox, IEEE Trans. Networkand Service Management, vol. 4, no. 1, June2007, pp. 5161, Project website (including GEANT/AbileneNetwork Topology and Traffic Dataset):

  9. Yin Zhang,MatthewRoughan Walter Willinger§ LiliQiu,The University of Texas at Austin University of Adelaide §AT&T Labs Research," Spatio-Temporal Compressive Sensing and Internet Traffic Matrices"

  10. HenrikAbrahamsson, Anders Gunnar, "Traffic Engineering in Ambient Networks: Challenges andApproaches" ,fhenrik,aegg@sics.seSwedish Institute of Computer ScienceBox 1263, SE-164 29 Kista, Sweden.

Leave a Reply