Performance Comparison of IP_MPLS and ATM

Embed Size (px)

Citation preview

  • 8/13/2019 Performance Comparison of IP_MPLS and ATM

    1/5

    Performance Comparison of IP, MPLS and ATM

    Based Network Cores using OPNET

    Hafiz M. Asif1and Md. Golam Kaosar

    2

    Department of Computer Engineering, King Fahd University of Petroleum and Minerals Dhahran 31261, [email protected], [email protected]

    Abstract - The core components of a network play vital role in

    its enhancement. For launching and providing Internet

    applications and services, vendors have come up with a

    variety of different technology based network cores and other

    components of Internet backbone. Among all the challenges

    for an efficient network core, traffic-routing is the most

    important. Both the routing algorithms and the router-

    technology are supposed to perform efficiently. We find that

    there are a number of technologies that offer packet routing

    service of which IP, ATM and MPLS are prominent. Some

    of the routers are designed to provide delivery of specific kind

    of traffic. The percentage of real time traffic (voice and

    video) over Internet is rapidly increasing due to growing

    trend of using real-time traffic oriented applications. People

    want to talk, watch TV, do video conferencing etc over

    Internet. This kind of real time traffic requires extra care

    because of more delay sensitivity, minimum QoS

    requirements and limited bandwidth. ATM and MPLS have

    some especial features to support real time traffic. On the

    other hand, it is almost impossible to replace already existing

    IP based network and Internet. Therefore numerous new

    features have been introduced in IP technology to support

    real time traffic. Thus, we find enough motivation to

    compare these three technologies in terms of their routing

    capability based on different performance metrics using

    OPNET simulator. We find that ATM and MPLS

    outperform pure IP (without modification) in terms of delay

    and response time to the exposed data.

    I INTRODUCTION

    The core components of a network play key role as far

    as overall network performance is concerned and it has

    nothing to do with end systems. The technology used for

    routing can also make significant difference. Currently,

    we have three main technologies for routing used in

    network cores: IP, ATM and MPLS based routing. IP is

    the oldest and highly used in network cores and a lot has

    been done and still research is going on it for further

    improvement. In order to enhance IP performance,

    various modifications to the routing techniques were

    proposed [1]. Similarly ATM and MPLS implementations

    were also carried out [2-4]. Moreover, the hybrid

    approach has also been studied in [5-7]. IP is still leading

    among them due to its pre-diverse usage and has grown

    relative features with which ATM and MPLS tried to

    dominate. Along with network core optimization problem

    in general, researchers are facing another challenge due to

    rapid increase in the usage of multimedia traffic

    (voice/video) over the Internet. The goal of this paper is

    to analyze the performance comparison of IP, ATM and

    MPLS based network cores when exposed to multimedia

    traffic. Our simulation study shows that MPLS and ATM

    outperform IP with respect to most of the selected

    performance metrics.

    The remainder of the paper is organized as follows. In

    the next section, we briefly review three routers / switches

    based on certain key characteristics. In Section III, we

    describe the simulation model and traffic scenarios. We

    then present the simulation results and evaluate the

    performance of different queuing disciplines in Section IV.

    Finally, we conclude our paper in Section V.

    II NETWORK CORE

    A network core is backbone of the entire network into

    which other personal or enterprise networks inject data.

    Design and capacity of the network core must be robust to

    serve all the connected subnets or independent nodesefficiently. Therefore, the core usually consists of high

    capacity links and efficient routing devices. We consider

    IP, ATM and MPLS based network cores in our work to

    see their effect on the network performance. IP based

    network core is designed to give efficient service based on

    certain features like best effort delivery, dynamic path

    establishment etc. Such a core shows poor performance at

    the time of congestion. On the other hand, ATM and

    MPLS tackle this problem in an efficient manner by

    making use of virtual path establishment and layered

    architecture approach. Owing to more careful strategies

    about delay and efficient use of available bandwidth, these

    cores are well suited for multimedia traffic. The key ideaof enhanced performance of ATM is its well-defined

    structure with exact degree of description about payload

    and other packet format. It is a cell based network

    protocol which encodes data traffic into fixed size cells.

    The frame size is 53 bytes where 48 bytes are data and the

    rest are for header information. It is, unlike IP, a

    connection-oriented technology, in which a connection is

    established between the two endpoints before the transfer

    of actual data starts. ATM supports several classes of

  • 8/13/2019 Performance Comparison of IP_MPLS and ATM

    2/5

    service to provide prioritization to certain type of data

    such as voice, video etc. Therefore, as mentioned earlier,

    it ensures QoS to real time traffic. However, this

    improvement is achieved in IP by making use of very high

    speed links available today (but we do not take this

    improvement into consideration for the sake of

    comparison). Similar to ATM, Multiprotocol Label

    Switching (MPLS) has been evolved to ensure some

    solutions to network problems like speed, scalability,

    traffic engineering and quality of service management with

    architecture different to that of ATM. More details about

    these technologies can be found in [8, 9].

    III METHODOLOGY AND APPROACH

    A Simulation Model

    We used OPNET Simulator to compare the three

    network cores. OPNET is a real time simulator

    specifically designed for network design and analysis. For

    our experiment, we consider a hypothetical network

    model. The core of the network consists of a number ofrouters in each of the three cases (ATM, IP, and MPLS).

    There are three types of traffic: FTP, VOICE and VIDEO

    on client sides with three corresponding servers on the

    receiving end. Table 1 shows the typical values and

    parameters of our traffic model. Note that the given data

    rates are just a typical value for one run which are varied

    to observe the performance metrics.

    In order to demonstrate the performance of the core

    consisting of a particular kind of routers in each of three

    scenarios, we considered a hypothetical network topology

    as illustrated in Fig. 1. The core of the network consists

    of routers of ATM, IP, and MPLS for each single scenario.

    All clients (FTP, VOICE and VIDEO) were connected tothe leftmost edge router whereas the single server was

    connected to the right edge router.

    Fig. 1 Network Topology for Simulation

    TABLE 1Link Speed 10Mbps

    Average FTP file size 5000 Bytes

    FTP data rate 100kbps

    Packet Size 2KByte

    Video frame rate 10 fps

    Video frame size 128x144 pixels

    Video data rate 56kbps

    Voice data rate 8kbps

    Voice codec G.729

    Video codec H.263

    For FTP traffic, we used exponential distribution for

    packet arrival, constant packet size and best-effort type of

    service. We use low resolution video starting at 10 fps

    (frames per sec) arrival rate and 128x120 pixels and keep

    increasing this rate and size. The ToS is Streaming

    Multimedia. For voice traffic, the voice encoder scheme is

    G.729, the silence and talk spurt lengths are exponentially

    distributed and ToS is Interactive Voice. All these settings

    were made using OPNET Application Attributes Profile.

    There are basically five types of delay that contribute to

    the overall end-to-end delay i.e. forwarding, propagation,

    queuing, packetization, and serialization. Forwarding

    delay is technology dependent and is measured in tens or

    hundreds of seconds (0.000001 sec). Propagation delay is

    trivial to calculate provided link speed and packet size are

    given. Serialization delay is the time it takes to place the

    bits of a packet onto the link when a router transmits apacket. It depends on packet size and link speed and

    typical value is around 1 msec [10]. Queuing delay is the

    time a packet has to wait in a queue and it can vary over

    time between zero seconds for a noncongested link, to the

    sum of the times that it takes to transmit each of the other

    packets that are queued ahead of it [10]. For Cisco

    routers, its typical value is around 8 msec for real time

    traffic.

    B Performance Metrics

    We use the following metrics for comparison and

    analysis purpose:

    Delay (sec) represents end-to-end delay. It is thesummation of all delays discussed in previous

    section.

    Throughput (pps) represents the total number ofpackets forwarded to higher layers per second.

    Utilization represents percentage of theconsumption of an available channel bandwidth

    where a value of 100.0 would indicate full usage.

  • 8/13/2019 Performance Comparison of IP_MPLS and ATM

    3/5

    FTP Download Response Time (sec) is the timebetween the request for channel is made and

    when the system starts download file.

    Normalized Delivered TrafficIt is the number ofcorrectly received packets normalized by sent

    packets from all clients. It indirectly shows the

    percentage of the dropped traffic over the

    network during simulation.

    For analysis purpose, we set the maximum queue size to

    be 500 packets.

    IV SIMULATION RESULTS AND ANALYSIS

    A Throughput

    Throughput depends on link speed and nature of the

    technology being used to transmit the data. We can see

    from the Fig. 2 that the throughput increases linearly with

    load until the channel gets saturated. Afterwards, it almost

    remains almost constant in cases of ATM and MPLS but

    there is an observable decrease in the case of IP due toboth its connectionless nature and heavy packet drop that

    may be because of congestion. Moreover, IP does produce

    relatively less throughput on account of its connectionless

    and best effort service characteristics. On the other hand,

    there is a virtual path defined for each packet in ATM and

    MPLS based network cores (each router / switch) which is

    responsible for the reduction of unreliability factor and

    thus improvement in overall network performance.

    0

    10

    20

    30

    40

    50

    60

    10 20 30 40 50 100 200

    Load (kpps)

    Throughput(kpps)

    IPMPLSATM

    Fig. 2 Throughput of the Network

    B Delay

    The absence of virtual connection is the key reason of

    relatively larger delay in IP based cores especially when

    compared with ATM, which has been clearly depicted in

    Fig. 3. MPLS falls between these two although its

    behavior is similar to that of ATM. This, in turn, verifies

    our earlier conclusion that ATM is supposed to be fit for

    real time traffic due to its prior path establishment feature.

    Finally, the quality of the real time traffic, especially for

    voice data, is deteriorated on such a high value of the

    delay.

    0

    150

    300

    450

    600

    750

    20 30 40 50 100 200

    Load

    Delay(msec)

    IP

    M PLSATM

    Fig. 3 Average End-to-End Delay of the Network

    C Utilization

    This is one of the key metrics to check the efficiency of

    the technology. This weighs how efficiently the

    technology makes use of the available resources such as

    bandwidth etc.

    The general trend of the utilization is to linearly

    increase with load till the point of saturation is reached(Fig. 4). Note that IP relatively makes good use of the

    0

    0.2

    0.4

    0.6

    0.8

    1

    1.2

    20 30 40 50 100 200

    Load

    Utilization

    IPMPLSATM

    Fig. 4 Utilization of the Network versus Load

  • 8/13/2019 Performance Comparison of IP_MPLS and ATM

    4/5

    available bandwidth although it produces lower

    throughput. This is because number of packets is large in

    case of IP compared to other two technologies because of

    no prior path establishment time. Moreover, IP router

    needs extra bandwidth in order to run routing algorithm

    several times to predict the best available path. Relatively

    less utilization of ATM and MPLS does not mean

    inefficiency. Instead it will benefit at the time of

    congestion and overloading situations.

    D FTP Download Response Time

    As mentioned earlier in section 2, ATM gives best

    response to all kinds of traffic. It means that there is little

    delay required to initiate the conversation and to let the

    client start sending its data. As shown in Fig. 5, FTP

    response time of ATM is considerably less compared to

    other two technologies. IP based core shows the worst

    behavior because of its random path detection (there is no

    virtual circuit in IP). This feature is very useful and can be

    exploited well for real time traffic such as voice and video.

    This directly takes us to the conclusion that ATM givesbetter performance for real time traffic as far any type of

    delay is concerned.

    0

    5

    10

    15

    20

    25

    30

    35

    20 30 40 50 100 200

    Load

    Responsetime(msec)

    IPMPLSATM

    Fig. 5 FTP Download Response Time for IP, ATM and MPLS

    E Normalized Traffic ReceivedWe can see the effect of connectionless and connection-

    oriented features of IP and other two technologies

    respectively by examining Fig. 6. Connectionless behavior

    of IP is responsible for more drop at the time of congestion

    (e.g. at 50kpps). However, it is not always the case i.e. IP

    device tries to avoid congestion by running congestion

    avoidance mechanism which most of the time works

    efficiently. Therefore, we can see from Fig. 6 that there is

    only one short interval of time where the received traffic

    normalized by traffic sent is significantly low. On the

    other hand, we make full use of connection-oriented nature

    of other two technologies thereby reducing dropped traffic

    to almost nil. Moreover, it is important to note that there

    is a gradual decrease in the received traffic common to all

    network cores with the increase in offered load due to

    capacity limit, network environment and some other

    reasons e.g. poor performance at physical layer etc. It is

    also important to mention that the current IP based

    network cores do not depict such a poor performance on

    account of improved IPv6 architecture and some other

    technology advancement which we did not include in our

    simulation. As mentioned earlier, we compare IPv4 versus

    ATM and MPLS because we also do not consider the

    recent improvements made in ATM and MPLS.

    0.5

    0.6

    0.7

    0.8

    0.9

    1

    1.1

    20 30 40 50 100 200

    Load

    ReceivedTraffic

    IP

    MPLS

    ATM

    Fig. 6 Normalized Received Traffic Comparison of IP, MPLS andATM technologies

    V CONCLUSIONS

    The structure of ATM and IP is entirely different (we

    used IPv4 in our simulation). ATM virtually establishes

    route before the transmission starts whereas IP randomly

    (or on demand) establishes routes for the packets to be

    transmitted by running routing algorithm (Dijkstra, DV

    etc.). This feature makes a subtle difference in the

    performance depicted by these two protocols. Dominanceand outclass performance of ATM over IP with respect to

    end to end delay, throughput and download response time

    clearly shows that prior establishment has a lot of

    advantages. One of the reasons of seldom use of ATM for

    Internet cores is because of the difficulty to replace the

    existing IP infrastructure which is widely spread all over

    the Internet. Moreover, the enhancement of IPv6 has also

    made possible to overcome the demerits that we showed in

    our simulation due to IPv4 structure.

  • 8/13/2019 Performance Comparison of IP_MPLS and ATM

    5/5

    ACKNOWLEDGEMENTS

    The authors would like to thank KFUPM (King Fahd

    University of Petroleum & Minerals) for its support.

    Special thanks must go to Dr. Abdul Waheed for his

    valuable pieces of advice during the course of simulation.

    Finally, we would also like to thank Muzibur Rehman

    working at KFUPM for his contribution in early survey

    work of this paper.

    REFERENCES

    [1] S. H. Lim, M. H. Yaacob, K. K. Phang, T. C. Ling,

    Traffic Engineering Enhancement to Qos-OSPF in

    Diffserv and MPLS Networks,IEE Proceedings online

    Vol. 151, No. 1, February 2004.

    [2] S. Crosby. Some Aspects of ATM Network

    Performance,IEE. Savoy Place, London WCPR OBL,

    UK , 1995.

    [3] L. Zhenyu, A Novel QoS Routing Scheme for MPLS

    Traffic Engineering, International Conference on

    Communication Technology, ICCT. April 2003, Vol. 1

    pp. 474-477.[4] Yuen, M. Ching; Jia, Weijia, Cheung, C. Chung,

    Efficient Distributed QoS Routing Protocol for MPLS

    Networks, 11th International Conference on Parallel

    and Distributed Systems Workshops, ICPADS 2005,

    July 2005, pp. 342-348.

    [5] Willis, Peter, Core IP and MPLS Networks, IEE

    Annual Course on Telecoms Networks - The Next

    Generations, London, United Kingdom, pp. 15-33, July

    2005.

    [6] Fujikawa, Fuyuki, Kuwabara, Kengo; Koda, Yoshiyuki;

    Kiuchi, Mai, Examination of Electric Power Utility

    Network Applying IP Router/MPLS Router/Wide-area

    Ethernet, IEEE Power Engineering Society General

    Meeting, Denver, USA, June 2004, pp. 901-904.[7] AlWehaibi, Abdullah, Kadoch, Michael, ElHakeem,

    A.hmed, Residual Multicast Loss for ARQ/diffserv

    over IP and MPLS Homogeneous Networks,

    Proceedings of the IASTED International Conference

    on Modelling, Sumulation and Optimization, Banff,

    Alta., Canada, July 2003 pp. 173-178.

    [8] L.A DaSilva, J.B Evans, D. Niehaus, V.S. Frost, R.

    Jonkman, Beng Ong Lee, G.Y. Lazarou, ATM WAN

    Performance Tools, Experiments, and Results, IEEE

    Communications Magazine, Vol. 35, No. 8, August

    1997, pp. 118-125.

    [9] Cisco Systems Inc., Internetworking Technologies

    Handbook: An Essential reference for every network

    professional, 4

    th

    Edition, September 2003.[10] Chuck Semeria, John W. Stewart III, Supporting

    Differentiated Service Classes in Large IP Networks,

    white paper from Juniper Networks, Inc. 2000