Source Routing Algorithms for Networks with Advance Reservations

Similar documents
A QoS Network Management System for Robust and Reliable Multimedia Services

QoS Routing Extensions to OSPF

Optimization of the Bandwidth Distribution Scheme for Handling Topology Changes

A LOAD-SENSITIVE QOS ROUTING ALGORITHM IN BEST-EFFORT ENVIRONMENT

A QoS Control Method Cooperating with a Dynamic Load Balancing Mechanism

HSM: A Hybrid Streaming Mechanism for Delay-tolerant Multimedia Applications Annanda Th. Rath 1 ), Saraswathi Krithivasan 2 ), Sridhar Iyer 3 )

DCRoute: Speeding up Inter-Datacenter Traffic Allocation while Guaranteeing Deadlines

Quality of Service Routing

ADAPTIVE AND DYNAMIC LOAD BALANCING METHODOLOGIES FOR DISTRIBUTED ENVIRONMENT

An OSPF Based Load Sensitive QoS Routing Algorithm using Alternate Paths

An Edge Router Based Protocol for Fault Tolerant Handling of Advance Reservations

An Efficient Rerouting Scheme for MPLS-Based Recovery and Its Performance Evaluation

CS 218- QoS Routing + CAC Fall 2003

Improving the usage of Network Resources using MPLS Traffic Engineering (TE)

Low complexity bandwidth guaranteed routing algorithms using path holding time

CHAPTER 3 EFFECTIVE ADMISSION CONTROL MECHANISM IN WIRELESS MESH NETWORKS

MPLS/RSVP-TE-BASED FUTURE UMTS RADIO ACCESS NETWORK

OPTIMAL MULTI-CHANNEL ASSIGNMENTS IN VEHICULAR AD-HOC NETWORKS

Branch-and-Bound Algorithms for Constrained Paths and Path Pairs and Their Application to Transparent WDM Networks

Classification and Evaluation of Constraint-Based Routing Algorithms for MPLS Traffic Engineering

Distributed Load-Sensitive Routing for Computationally-Constrained Flows

Admission Control in Time-Slotted Multihop Mobile Networks

Overlay Networks for Multimedia Contents Distribution

Performance Evaluation of Mesh - Based Multicast Routing Protocols in MANET s

Evaluation of Information Dissemination Characteristics in a PTS VANET

QoS-Aware IPTV Routing Algorithms

LSP Setup Arrival Reordering Approach for MPLS-TE Routing

International Journal of Scientific & Engineering Research, Volume 7, Issue 12, December ISSN

CSMA based Medium Access Control for Wireless Sensor Network

Implementation and simulation of OLSR protocol with QoS in Ad Hoc Networks

DEPARTMENT of. Computer & Information Science & Engineering

Prioritization scheme for QoS in IEEE e WLAN

CHAPTER 5 ANT-FUZZY META HEURISTIC GENETIC SENSOR NETWORK SYSTEM FOR MULTI - SINK AGGREGATED DATA TRANSMISSION

MPLS-TE Configuration Application

Presented by: B. Dasarathy OMG Real-Time and Embedded Systems Workshop, Reston, VA, July 2004

A Time-To-Live Based Reservation Algorithm on Fully Decentralized Resource Discovery in Grid Computing

Computation of Multiple Node Disjoint Paths

' INRIA Rocquencourt, Domaine de Voluceau

Lecture 9. Quality of Service in ad hoc wireless networks

Projected node. Pruned node. h h + 1

Factors Affecting the Performance of Ad Hoc Networks

FAULT TOLERANT REAL TIME COMMUNICATION IN MULTIHOP NETWORKS USING SEGMENTED BACKUP

Project Report: QoS Enhancement for Real-Time Traffic in IEEE WLAN

Providing Quality of Service Guarantees Using. Only Edge Routers

Quality of Service in the Internet

Internet Quality of Service: an Overview

Performance Evaluation of Active Route Time-Out parameter in Ad-hoc On Demand Distance Vector (AODV)

A FORWARDING CACHE VLAN PROTOCOL (FCVP) IN WIRELESS NETWORKS

THE EFFICIENCY OF CONSTRAINT BASED ROUTING IN MPLS NETWORKS

Quality of Service in the Internet

Performance Analysis of Storage-Based Routing for Circuit-Switched Networks [1]

Networking Quality of service

Evaluation of Performance for Optimized Routing in MPLS Network

Practical QoS Network System with Fault Tolerance

Comparative Evaluation of Adaptive Price-based Admission Control Algorithms for Bandwidth Allocation

Effective Bandwidth Allocation for WiMAX Mesh Network

CSCD 433/533 Advanced Networks Spring Lecture 22 Quality of Service

Delayed reservation decision in optical burst switching networks with optical buffers

Achieving Distributed Buffering in Multi-path Routing using Fair Allocation

A Congestion Contribution-based Traffic Engineering Scheme using Software-Defined Networking

Distributed QoS Routing for Backbone Overlay Networks

Basic Idea. Routing. Example. Routing by the Network

On-Line Routing in WDM-TDM Switched Optical Mesh Networks

Bayeux: An Architecture for Scalable and Fault Tolerant Wide area Data Dissemination

Cost-based Pricing for Multicast Streaming Services

Keywords Minimum Spanning Tree, Mobile Adhoc Network (MANET), Multicast, Overhead, Scalability, Spanning Tree.

Routing by the Network

Boosting the Performance of Myrinet Networks

Implementing QOS Policy in MPLS Network

On Optimal End-to-End QoS Budget Partitioning in Network Dimensioning

Power aware Multi-path Routing Protocol for MANETS

QoS Routing and Traffic Scheduling in Long-Distance Wireless Mesh Networks

Simulation Analysis of Linear Programming Based Load Balancing Algorithms for Routers

IP Differentiated Services

Influence of Link State Update Algorithms on the Cost of QoS Path Computations

Performance of Multihop Communications Using Logical Topologies on Optical Torus Networks

1 Multipath Node-Disjoint Routing with Backup List Based on the AODV Protocol

Advance and Immediate Request Admission: A Preemptable Service Definition for Bandwidth Brokers

Designing A New Routing Simulator for DiffServ MPLS Networks *

The Performance of MANET Routing Protocols for Scalable Video Communication

Efficient QoS Partition and Routing in Multiservice IP Networks

A Fuzzy System for Adaptive Network Routing

On the Scalability-Performance Tradeoffs in MPLS and IP Routing

Congestion Control in Mobile Ad-Hoc Networks

Quality of Service in the Internet. QoS Parameters. Keeping the QoS. Leaky Bucket Algorithm

A host selection model for a distributed bandwidth broker

Performance Analysis of Proactive and Reactive Routing Protocols for QOS in MANET through OLSR & AODV

Lecture 17 Multimedia Transport Subsystem (Part 3)

Module 15: Network Structures

G BAM: A Generalized Bandwidth Allocation Model for IP/MPLS/DS-TE Networks

arxiv: v2 [cs.ni] 23 May 2016

Modelling direct application to network bandwidth provisioning for high demanding research applications

Precomputation Schemes for QoS Routing

The Virtual Resource Manager: An Architecture for SLA-aware Resource Management

Advanced Fuzzy Class-Based Routing in MPLS-TE Networks

QoS-Aware Hierarchical Multicast Routing on Next Generation Internetworks

Delay Performance of Multi-hop Wireless Sensor Networks With Mobile Sinks

A Network Optimization Model for Multi-Layer IP/MPLS over OTN/DWDM Networks

Heuristic Algorithms for Multiconstrained Quality-of-Service Routing

Quality of Service Routing. Anunay Tiwari Anirudha Sahoo

Transcription:

Source Routing Algorithms for Networks with Advance Reservations Lars-Olof Burchard Communication and Operating Systems Technische Universitaet Berlin ISSN 1436-9915 No. 2003-3 February, 2003

Abstract In contrast to immediate reservations where resources are allocated in a just-intime manner, advance reservations which allow to reserve network bandwidth long before it is actually required are an opportunity to provide enhanced functionality in a computer network. In such an environment, source routing algorithms can be applied using an MPLS based network infrastructure with bandwidth brokers as control instance. In recent publications, the precomputation of a set of k paths has been proposed as a suitable mechanism to facilitate fault tolerance mechanisms in advance reservation environments. However, the impact of these routing algorithms on the network performance e.g. in terms of call acceptance rate was not examined. In this paper, the multiple path routing strategies are evaluated under different conditions. It can be shown that the performance is always better compared to other source routing strategies. This allows to use the algorithms not only for supporting link resilience mechanisms but also as default routing strategy in advance reservation environments. In addition to the routing algorithms, a strategy called path switching is introduced, which switches flows to alternative paths during their transmission time and thus increases the overall performance of the network.

Chapter 1 Introduction In general, two types of resource reservations in networks can be distinguished: immediate reservations which are made in a just-in-time manner and advance reservations which allow to reserve resources a long time before they are actually used. In contrast to the field of immediate reservations in computer networks, so far advance reservations have not been in the focus of many researchers. In current IP networks such as the Internet, guarantees cannot be given concerning the stability of the computed routes. One of the main reasons is unpredictable routing behavior. Changed routes cannot be foreseen which results in a serious problem for reliable admission control. However, advance reservations are useful in any application where large amounts of data have to be transmitted over a network and the time of such a transmission is known in advance. This is the case e.g. in web caching or distributed multimedia applications [2] where large amounts of content such as video files have to be transmitted up to a certain, predefined deadline. Another example is grid computing, where typical computations on the distributed parallel systems result in large amounts of data which also have to be transmitted in time between the different machines. Bandwidth Broker MPLS Domain Figure 1.1: MPLS network with bandwidth broker 1

The multi protocol label switching (MPLS) architecture is an interesting approach for supporting advance reservations in computer networks. MPLS provides mechanisms for traffic engineering and in particular explicit routing which is important and makes MPLS ideally suited to support advance reservations. We consider an architecture using a bandwidth broker as network management system, i.e. the broker grants and denies access to the network, and initiates the explicit routing process. The architecture is outlined in Figure 1.1. In such an environment, admission control for advance reservations is performed such that a route with sufficient bandwidth is determined and set up by the bandwidth broker using the explicit routing functionality of MPLS. Using such an architecture allows to set up advance reservation services on top of existing network infrastructure and provides the opportunity for the implementation of advance reservations in future networks. In this document, source routing algorithms for such environments are presented. The motivation for the developments presented here is that previous work showed routing strategies that distribute flows onto multiple paths are a useful technique to implement failure handling mechanisms in networks. This holds for immediate reservations [7] but is also applicable in advance reservation environments [1]. Such strategies compute a set of k paths for each two end points of the network and routes each flow using a path from these sets. In case of a link failure, the remaining k 1pathscanbeusedtoreroutetheflows affected by the failure. Such strategies have apparent advantages, e.g. that path computation can be performed at start time of the system and admission control for each flows only requires to check the k paths for sufficient bandwidth. In this document these algorithms are examined regarding their suitability for routing flows in advance reservation environments in general, i.e. their performance compared to other routing strategies is examined which so far has not been done in detail. For that purpose, routing strategies based on Dijkstra s shortest path (DSP) algorithm are derived from QoS routing algorithms used in the field of immediate reservations and compared to the k path algorithms. The DSP based strategies were adapted to the advance reservation environment and several variants were implemented which used different metrics to rate the available paths. These algorithms compute paths on-demand individually for each requested flow. In addition to the on-demand and precomputed multiple path strategies, the advance reservation environment allows to switch flows to different paths in case no single path with sufficient bandwidth is available during the whole transmission period. This strategy can be used in addition to the routing algorithms and further increases the performance of the network. The performance of the k path algorithms is better than other source routing algorithms that compute paths individually for each flow. The advantages of the k path algorithms, i.e. the suitability for fault tolerance and the simplified admission control process, together with similar or even better performance lead to the conclusion that these strategies are well suited to be used for routing in advance reservation environments. In the following sections, after discussing related work we describe the two 2

basic types of algorithms for calculating paths, i.e. the individual path computation based on DSP and the k path strategies. Following that, the simulation environment is described together with detailed results of the experiments. The paper concludes with some final remarks and a discussion of future work. 1.1 Related Work Early works considering advance reservations concentrated on the basic requirements to facilitate such mechanisms [4], while others developed architectures such as an agent based approach for enabling a scalable advance reservation mechanism [10] based on OSPF routing infrastructure. In contrast to those papers, we assume to use an MPLS [9] aware core network with a bandwidth broker responsible for admission control and path setup. Others extended the advance reservation environment to incorporate not only reservations in networks but also reservations on the computer that are involved in the transmissions [11]. Although their paper only discussed the requirement to reserve resources in scenarios like video streaming, this is also an important aspect in the field of grid computing. An example for an advance reservation system designed to support grid computing can be found in [5]. Other applications for advance reservations are from other fields of distributed computing such as distributed video server systems as described in [2], where the timely transmission of large media files requires support for reserving the network resources in advance. Routing algorithms for supporting advance reservations so far have been examined only in [6]. In contrast to our approach, in [6] only the relation between advance reservations and routing in terms of computational complexity is discussed. The focus of the considerations is to examine routing algorithms for different problems related to advance reservations such as searching for the earliest start time for a transmission of given bandwidth and duration. The basic problem of finding a suitable path for a transmission as examined in our paper is solvable in polynomial time. The algorithms for computing k paths examined here are based on the considerations in [1] and [7]. In [1] two k path algorithms were compared with respect to their suitability to support fault tolerance in advance reservation environments. The first algorithm was derived from the maximally disjoint path algorithm presented in [7], designed to be used in immediate reservation environments. The other algorithm examined in [1] was taken from [3] and simply computes the k shortest paths. The comparison both strategies showed, that the performance of the k shortest path algorithm was far better. The reason was the limited path set computed by the maximally disjoint algorithm. Therefore, only the k shortest approach was examined here. The basic strategy was adapted to support advance reservations and extended using well-known metrics such as Shortest-Widest [8]. 3

Chapter 2 Routing Algorithms In this section, we describe the different routing algorithms and their properties. At first, the admission control procedure using individual and k path algorithms is presented. 2.1 Admission Control The bandwidth broker processes requests for network quality-of-service from clients. The QoS guarantee in our case is only bandwidth [6, 10]. bandwidth request slot time Figure 2.1: Advance reservations: slotted time and status storage In order to perform reliable admission control, it is important to keep status about future utilization for each link. This means, the allocated bandwidth at each point in time is recorded. The time interval for which requests can be issued is restricted and the interval is called book-ahead time. Whenarequestis issued, the management system checks whether sufficient bandwidth is available during the requested duration. As in [6, 10], we use slotted time which means the book-ahead time is divided into slots of fixed size, e.g. minutes (see Figure 2.1). Hence, requests are issued for a consecutive number of slots which means the start and stop times are also submitted with each request. Status, i.e. the allocated bandwidth, is kept for each slot on each link of the network. The restriction of the book-ahead time reduces the amount of data to be stored. During admission control, links can only be considered to route a given flow if 4

sufficient bandwidth is available for each of the requested slots. Admission control in the case of individual path calculation means to determine a path between source and sink with sufficient bandwidth during the whole time interval between start and stop time of the transmission. When k path strategies are applied, a set of paths is precomputed for each pair of nodes in the network, e.g. at start time of the management system. The admission control process then only requires selecting one of the precomputed paths with sufficient bandwidth. In case several paths provide sufficient bandwidth one of the k paths is chosen according to the strategies described in Section 2.2. The details of the path computation and path selection process are described in the following sections. 2.2 Path Computation 2.2.1 On-Demand Path Computation In this section, the individual path computation strategies for computing paths on-demand are described. The basic idea of these algorithms is to compute a feasible path for each request and to use other than shortest path metric, e.g. the widest path or the path with the maximally available bandwidth. The algorithms are based on Dijkstra s shortest path (DSP) algorithm and called Advance Reservation Shortest Path (ASP) algorithms. These strategies always find a feasible path if one exists. The basic idea is to compute the link weights for the original DSP algorithm using information about the link utilization during the requested transmission interval. The different metrics are derived from QoS routing algorithms for immediate reservations and applied to the advance reservation environment. ASP (G(V,E),s,d,bw,t start,t stop ) 1 for each v V do 2 w[v] = 3 φ[v] =NIL 4 T = V 5 w[s] =0 6 7 while T 0do 8 u = get min(t ) 9 for each v T Adj[u] do 10 if ((w[u]+weight(u, v, t start,t stop ) <w[v])and 11 (c(u, v, t start,t stop ) >bw)) 13 then 12 w[v] =w[u]+weight(u, v, t start,t stop ) 13 φ[v] =u The ASP algorithm has the same complexity as the original DSP algorithm, 5

i.e. O(E log(v )). In the algorithm, the array w contains the current node weights and φ is the array storing the actual path, i.e. predecessors of nodes. c(u, v, t start,t stop ) denotes the available capacity on link (u, v) during the period from t start and t stop. The function get min determines the node u T with the minimum weight d(u) and removes it from T. The condition in line 11 assures, that sufficient bandwidth is available on the computed path. The function weight(u, v, t start,t stop ) reflects the load on link (u, v) during the requested transmission interval [t start,t stop ]. In contrast to immediate reservation scenarios, advance reservation architectures as discussed here allow to use information about the link load during the whole transmission period. In case, the function weight always equals 1, the algorithm presented above computes the shortest available path (ASP-Shortest). Two other variants were evaluated, using different implementations of the function weight. ASP-Shortest: uses the hop count as path metric, i.e. the weight equals 1 for each link ASP-Widest: computes the widest available path. The algorithms uses the maximal available bandwidth (computed over all slots within the requested transmission interval) as link weight. ASP-MinLoad: uses the average load during the whole requested transmission period as link weight. Let c(φ, t) denote the allocated bandwidth at slot t on link φ, then the average load on φ during the interval [t 1,t n ]is i n c(φ,ti) i=1 computed as load(φ, t) = n. The link weight is then computed as weight(φ, t start,t stop )=1+load(φ, t). The ASP-Widest and ASP-MinLoad strategies do not simply compute the widest respectively the path with minimal load, but by using the respective metric as link weight also include the path length into the weight for the whole path. This prevents the algorithms from using paths which are too long and does not require to take additional measures to additionally include the hop count into the weight of a path. Correctness of the Algorithms In order to work correctly, i.e. to reliably compute the shortest available path according to the selected metric, the previously presented ASP algorithm requires the weight function to produce only non-negative values. This is obviously the case for any of the presented variants. 2.2.2 Multiple Path Computation The same metrics as chosen for the on-deamnd path computation can be applied to the k path strategies, i.e. a path from the precomputed set is chosen using the following metrics: 6

KP-Shortest-Random: uses the shortest feasible path from the path set. In case, several paths have the same hop count, one is randomly selected. KP-Shortest-Widest: similar to KP-Shortest-Random, but if several paths with equal length exists, uses the widest among them. KP-Shortest-MinLoad: in case several paths with equal weight exists this strategy uses the path with minimal average load, compute as described for ASP-MinLoad. KP-Widest-Shortest: this algorithm uses the path with the maximal available bandwidth during the requested transmission period. In case several equally weighted feasible paths exist, the shortest is selected. 2.3 Path Switching The advance reservation scenario provides the opportunity to increase the performance by allowing to switch flows to a different path during the transmission time one or more times in case no path with sufficient bandwidth during the whole requested transmission period is available. This increases the call acceptance rate and is completely transparent for client applications since it does not require additional efforts such as reordering packets as is the case for the parallel transmission of packets on multiple paths. Instead, the network management is responsible for switching the flows onto a different path at a given point in time. In contrast to immediate reservations, where the environment does not allow to implement such functionality, it is possible to implement this technique as part of an advance reservation service. Admission control is implemented by computing the paths with the longest available intervals with sufficient bandwidth such that the whole transmission can be admitted. Path switching can be applied using any of the individual and multiple path routing strategies described above. 7

Chapter 3 Evaluation In this section, the different routing strategies previously presented are examined and compared. In order to determine the performance of the routing algorithms, using only the call acceptance rate is not sufficient since this does not reflect the amount of transmitted bytes of the admitted requests and hence strategies that prefer short calls perform better while the utilization of the network remains low. For this reason, the bandwidth blocking rate (bbr) was used as the second bandwidth(i) metric. It is defined as bbr = i R bandwidth(i), wherer is the set of rejected i A requests and A is the set of requests issued [8]. 3.0.1 Simulation Environment cost239 eqos Figure 3.1: Network topologies The algorithms were evaluated in a simulation environment. For that purpose, two network topologies as shown in Figure 3.1 were selected. Both represent realistic ISP backbone topologies. Each link was assigned a capacity of 100 MBit/s. Each simulation was run for a duration of 10000 slots. Within this period of time, requests were generated with exponentially distributed duration with 8

a mean of 100 slots. The time between the requests issued and the start of the transmission followed a log-normal distribution with a mean value of 250 slots. The bandwidth requirement was uniformly distributed between 64 KBit/s and 1 MBit/s. The load is evenly distributed on the network, i.e. the start and destination node for each flow is chosen randomly form all the available nodes following a uniform distribution. 3.0.2 Results In this section, the results of the simulations are presented. Firstly, the properties of the k path strategies are examined. load=low load=medium load=high 79 67 22,5 77 65 20,5 63 75 61 18,5 73 59 16,5 84 82 80 78 76 57 55 53 51 49 cost239 17 eqos 20 19 18 Shortest-Random Shortest-Widest Shortest-MinLoad Widest-Shortest Path-Switching Figure 3.2: Call acceptance rates depending on the parameter k The call acceptance rates as a function of the parameter k are depicted in Figure 3.2 for both topologies. In addition to the KP strategies presented in Section 2.2, the results of the Shortest-Widest strategy with the path switching option are given. It can be seen, that the network load is an important factor for the performance of the strategies. With low network load, the performance increases with rising k. When the load increases, the optimal performance is achieved with low values of k. With high network load the optimal performance is reached with k = 1 due to the fact that only the shortest path is used and hence, no bandwidth is wasted but can be used for requests with different end points. Although the results for the two topologies cannot be directly compared, it is obvious that the optimal value of k depends not only on the load but also on the topology. Using path switching allows to achieve an additional performance gain, especially in the medium load scenarios. The impact of k on the bandwidth blocking rate is shown in Figure 3.3. It can be observed that again Shortest-Widest achieved the best results - further increased by path switching - whereas Widest-Shortest performs drastically worse. The dependence of the bandwidth blocking rate on the network load is also apparent. In Figure 3.4, the results for individual and k path strategies are compared 9

load=low load=medium load=high 36 34 49 87 32 85 44 30 83 28 39 81 cost239 73 71 69 67 60 58 56 54 52 50 eqos 33 31 29 27 25 23 Shortest-Random Shortest-Widest Shortest-MinLoad Widest-Shortest Path-Switching Figure 3.3: Bandwidth blocking rate depending on the parameter k Path Switching KP-Shortest- Widest (k=10) ASP-MinLoad ASP-Widest ASP-Shortest 63 64 65 66 67 68 call acceptance rate (%) Path Switching KP-Shortest- Widest (k=10) ASP-MinLoad ASP-Widest ASP-Shortest 48 49 50 51 52 53 54 55 56 57 call acceptance rate (%) cost239 Path Switching KP-Shortest- Widest (k=10) ASP-MinLoad ASP-Widest ASP-Shortest eqos Path Switching KP-Shortest- Widest (k=10) ASP-MinLoad ASP-Widest ASP-Shortest 38,5 39 39,5 40 40,5 41 41,5 bandwidth blocking rate (%) 48 49 50 51 52 53 54 55 bandwidth blocking rate (%) Figure 3.4: Comparison of k path and individual path computation strategies in the medium load scenario using the most successful k path strategy KP- Shortest-Widest. In general, the results do not largely differ. However, it can be observed that using ASP-Widest - as for the k path algorithms - leads to the best performance. The most important result however is, that the KP- Shortest-Widest algorithm always achieves a better performance than any of the ASP strategies. Using the path switching approach in addition to KP- Shortest-Widest, up to 5% better performance can be achieved. 3.0.3 Path Switching Overhead With the path switching approach, additional administrative overhead is required for switching a flow to a different path. The number of path switches should therefore be kept low. 10

switched flows (%) switched flows (%) 20 15 10 5 0 20 15 10 5 0 # switches per flow (avg) cost239 # switches per flow (avg) 3,5 3 2,5 2 1,5 1 0,5 0 3,5 3 2,5 2 1,5 1 0,5 0 high medium low eqos Figure 3.5: Path switching overhead as a function of k In Figure 3.5, the percentage of flows that are switched at least once during their transmission time and the average amount of path switches for a single flow is depicted in dependence of the path set size k and the network load, i.e. low, medium, and high load. The diagrams show, that the number of flows that are switched rises with increasing load with the percentage of flows switched at least once ranging between 6 and 15 %. Other parameters such as the topology or the parameter k do not have a significant impact. 3.0.4 Unforced Reject Rate In the previous sections, the k path strategies were shown to reach the same and even better network performance than other path computation strategies. However, restricting the set of available paths can lead to situations were requests are rejected although sufficient bandwidth is available at the time the requests are issued. unforced reject rate (%) 20 15 10 5 cost239 0 unforced reject rate (%) eqos 50 40 30 20 10 0 low medium high Figure 3.6: Unforced reject rate for varying k and load In Figure 3.6, the unforced reject rate urr, denoting the percentage of flows 11

that are rejected although sufficient bandwidth was available at the time the requestswereissued, isdepicted for differentvalues ofk under low, medium, and high load. The actual routing strategy used for these results was KP-Shortest- Widest. The urr drops rapidly with increasing k. Fork = 10, in both topologies less than one percent of the requests are rejected although sufficient bandwidth was available. This amount is negligible and hence in practical implementations does not play a significant role. When applying path switching, urr can be further decreased, i.e. urr drops below 0.01% with k = 10 in both topologies. 12

Chapter 4 Conclusion In this document, source routing strategies for advance reservations were presented. Several individual and multiple path routing strategies were developed and compared. The architecture of the reservation system is based on networks with MPLS support and bandwidth brokers as management system. The motivation for the work was the succesful implementation of k path strategies in order to support fault tolerance mechanisms in advance reservation environments, however lacked an evaluation of the performance in terms of call acceptance rate and bandwidth blocking rate. The k path algorithm was compared to individual path computation strategies in several environments under different load conditions. In addition, the advance reservation scenario allows to switch flows to alternative paths during the duration of the transmission. This can be efficiently implemented using the k path algorithms and increases the performance by up to 5%. The most important result of the evaluations is that the performance of the k path strategies is comparable to that of other path computation algorithms. Moreover, in some environments even better performance can be achieved. This holds especially for evenly distributed load. This means, that the k path strategies are not only useful to support the implementation of fault tolerance mechanisms and reduce the admission speed of the admission control process, but are also competitive in terms of the network performance (call acceptance rate and bandwidth blocking rate). Therefore, these algorithms are well suited to be used for the admission control process. For this paper, not any possible QoS routing algorithm available for immediate reservations could be adapted and tested, therefore the different ASP variants may not reflect the whole variety of opportunities. However, the performance of the k path algorithms is based on the restriction of the path set in a way that under no circumstances paths are used that are too long which cannot be applied for the individual path computation in a similar way. Together with the flexibility to adapt the parameter k to the actual environment, i.e. topology and network load, this guarantees good performance using the proposed strategies. 13

Bibliography [1] L.-O. Burchard and M. Droste-Franke. Fault Tolerance in Networks with an Advance Reservation Service. In 11th International Workshop on Quality of Service (IWQoS), 2003. [2] L.-O. Burchard and R. Lüling. An Architecture for a Scalable Video-on- Demand Server Network with Quality-of-Service Guarantees. In Proceedings of the 5th Intl. Workshop on Distributed Multimedia Systems and Applications (IDMS), Lecture Notes in Computer Science, Springer, volume 1905, pages 132 143, 2000. [3] D. Eppstein. Finding the k Shortest Paths. SIAM Journal on Computing, 28(2), 1998. [4]D.Ferrari,A.Gupta,andG.Ventre. Distributed Advance Reservation of Real-Time Connections. In Network and Operating System Support for Digital Audio and Video, pages 16 27, 1995. [5] I. Foster, C. Kesselman, C. Lee, R. Lindell, K. Nahrstedt, and A. Roy. A Distributed Resource Management Architecture that Supports Advance Reservations and Co-Allocation. In 7th International Workshop on Quality of Service (IWQoS), 1999. [6] R. Guerin and A. Orda. Networks with Advance Reservations: The Routing Perspective. In Proceedings of IEEE INFOCOM, 2000. [7] S. Lee and M. Gerla. Fault-Tolerance and Load Balancing in QoS Provisioning with Multiple MPLS Paths. In Ninth International Workshop on Quality of Service (IWQoS), 2001. [8] Q. Ma and P. Steenkiste. On Path Selection for Traffic with Bandwidth Guarantees. In International Conference on Network Protocols (ICON), 1997. [9] E. Rosen, A. Viswanathan, and R. Callon. Multiprotocol Label Switching Architecture. ftp://ftp.isi.edu/in-notes/rfc3031.txt, January 2001. RFC 3031. 14

[10] O. Schelen and S. Pink. An Agent-based Architecture for Advance Reservations. In 22nd Annual Conference on Computer Networks (LCN 97), 1997. [11] L. C. Wolf and R. Steinmetz. Concepts for Resource Reservation in Advance. Multimedia Tools and Applications, 4(3):255 278, 1997. 15