Vous êtes sur la page 1sur 12

1320

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

VOL. 18,

NO. 9,

SEPTEMBER 2007

Bandwidth Management for Supporting Differentiated-Service-Aware Traffic Engineering


Tong Shan, Member, IEEE, and Oliver W.W. Yang, Senior Member, IEEE
AbstractThis paper presents a bandwidth management framework for the support of Differentiated-Service-aware Traffic Engineering (DS-TE) in multiprotocol label switching (MPLS) networks. Our bandwidth management framework contains both bandwidth allocation and preemption mechanisms in which the link bandwidth is managed in two dimensions: class type (CT) and preemption priority. We put forward a Max-Min bandwidth constraint model in which we propose a novel use it or lend it strategy. The new model is able to guarantee a minimum bandwidth for each CT without causing resource fragmentation. Furthermore, we design three new bandwidth preemption algorithms for three bandwidth constraint models, respectively. An extensive simulation study is carried out to evaluate the effectiveness of the bandwidth constraint models and preemption algorithms. When compared with the existing constraint models and preemption rules, the proposed Max-Min constraint model and preemption algorithms improve not only bandwidth efficiency, but also robustness and fairness. They achieve significant performance improvement for the well-behaving traffic classes in terms of bandwidth utilization and bandwidth blocking and preemption probability. We also provide guidelines for selecting different DS-TE bandwidth management mechanisms. Index TermsResource management, admission control, differentiated service, traffic engineering.

1 INTRODUCTION
DS-TE brings the following benefits over classical TE. First, by enforcing different bandwidth commitments for different service classes, DS-TE ensures that the amount of traffic of each class routed over a link matches the configuration of the link scheduler so that the appropriate QoS is provided to different service classes. Second, DS-TE simplifies the configuration of link schedulers and avoids frequent adaptive adjustment of the scheduling parameters. Third, by balancing the traffic load of different service classes on links, DS-TE prevents performance interference between the real-time and non-real-time traffic classes. In general, DS-TE consists of two major functions: bandwidth management and route computation. Although many studies have been conducted on TE, most of them focused on the route selection algorithms [19], [20], [21], and little effort has been put into DS-TE bandwidth management techniques. Obviously, when TE is performed in a DiffServ scenario, a bandwidth manager is indispensable for every network node to enforce different bandwidth constraints for each class, to perform separate admission controls for each class, and to flood the network with bandwidth availability information on a per-CT level. Designing a bandwidth manager for the support of TE in the DiffServ environment is what we focus on in this paper. There are more requirements for the DS-TE bandwidth management than for the conventional bandwidth management. First, the DS-TE bandwidth management must be able to enforce different bandwidth constraints for different CTs. Second, it must be able to process the bandwidth requests with at least three parameters: the requested bandwidth, the CT, and the priority. Third, it is desirable to support bandwidth preemption [1], [6]. Bandwidth preemption is becoming more useful in the DiffServ environment, where it can be used to assure highpriority traffic trunks with relatively favorable paths and
Published by the IEEE Computer Society

S an important application of multiprotocol label switching (MPLS), Traffic Engineering (TE) has attracted much attention for its ability to achieve end-to-end quality of service (QoS) [4], [5]. In the classical TE mechanism, bandwidth is managed on an aggregate basis. All packets toward the same destination are routed collectively according to the single constraint of available link bandwidth and will follow the same Label Switched Path (LSP). Since classical TE operates without referring to different classes of services, it may not be optimal in a Differentiated Service (DiffServ) environment [1], [2]. To address this issue, the Internet Engineering Task Force (IETF) proposed the DiffServ-aware TE (DS-TE), which performs TE at a perclass level [6]. Traffic flows toward a given destination can be transported on separate LSPs on the basis of service classes and may follow different paths. In the DS-TE solution, the IETF proposed to advertise the available bandwidth on a per-class type (CT) basis to improve the scalability of link-state advertisement, with a CT being a group of traffic trunks crossing a link that is governed by a specific set of bandwidth constraints [6]. The TE class is introduced in [6] as a pair of a CT and a preemption priority allowed for that CT.

. T. Shan is with Nortel Networks, Ottawa, Ontario, Canada K2H 8E9. E-mail: tshan@ieee.org. . O.W.W. Yang is with the School of Information Technology and Engineering, University of Ottawa, Ottawa, Ontario, Canada K1N 6N5. E-mail: yang@site.uottawa.ca. Manuscript received 23 June 2005; revised 13 Mar. 2006; accepted 12 Oct. 2006; published online 9 Jan. 2007. Recommended for acceptance by S. Das. For information on obtaining reprints of this article, please send e-mail to: tpds@computer.org, and reference IEEECS Log Number TPDS-0303-0605. Digital Object Identifier no. 10.1109/TPDS.2007.1052.
1045-9219/07/$25.00 2007 IEEE

SHAN AND YANG: BANDWIDTH MANAGEMENT FOR SUPPORTING DIFFERENTIATED-SERVICE-AWARE TRAFFIC ENGINEERING

1321

reliable resources. So far, the IETF only proposed the basic rules [4], [5] in which each connection is assigned with one holding priority for keeping resources and one setup priority for taking resources. A connection with a high setup priority is allowed to preempt a connection with a lower holding priority. For a system to be stable, the holding priority of a connection must be higher (numerically smaller) than its setup priority. Previous works on preemption [15], [16], [17], and [18] focused on connection preemption selection, that is, selecting an appropriate set of connections to be preempted. In these works, each connection is essentially associated with two parameters: bandwidth and priority. Without considering the CT together with the priority, their algorithms are not feasible for a full support of DS-TE. Besides, their algorithms did not provide the preemption fairness among CTs, nor did they protect the normal load or underloaded CTs from being overwhelmed by the overloaded CTs. Among the previous works [6], [7], [8], [9], [25] are the most closely related to this paper, which deals with the DS-TE bandwidth management. The Maximum Allocation Model (MAM) [7], the Russian Doll Model (RDM) [8], and the Maximum Allocation with Reservation (MAR) [25] are three IETF-proposed bandwidth constraint models for supporting DS-TE. The performances of MAM, RDM, and MAR have been evaluated in [9], [25], but the results are very preliminary. First, each CT is associated with a different priority. Such a one-to-one mapping implies that only interCT preemption was studied. Second, only the blocking/delay and blocking/preemption probabilities were evaluated, without the bandwidth efficiency being discussed. Bandwidth preemption was recognized as an important piece of DS-TE bandwidth management, but no preemption strategy was proposed. In summary, none of these previous works proposed a comprehensive solution to the DS-TE bandwidth management, with each focused on one or two aspects of the issue. To the best of our knowledge, this work is the most complete solution on the DS-TE bandwidth management. The Virtual Partitioning (VP) in [11], [12], [13] is a wellknown resource management scheme, which provides a fair, robust, and efficient way of resource sharing. At first glance, the Max-Min constraint model proposed in Section 2.3 might look similar to the VP, but it is actually a very different model. The VP is essentially a bandwidth protection mechanism, where each traffic class is allocated with a nominal amount of bandwidth on the basis of its traffic load forecast and QoS requirements. A traffic class is allowed to exceed its nominal allocation as long as there is a certain amount of protection bandwidth still unreserved. Hence, the VP cannot enforce different maximum or minimum bandwidth constraints for different classes of services. Besides, it does not deal with bandwidth preemption. Therefore, it is not feasible for the DS-TE solution. As we shall see in Section 2, both the maximum reservable bandwidth and the minimum guaranteed bandwidth of our proposed Max-Min constraint model are bandwidth constraints of a CT, which govern the bandwidth allocation of a group of LSPs belonging to the CT. This is quite different from the Max-Min rate allocation in asynchronous transfer mode (ATM) networks in [26], [27], where the minimum bandwidth guarantee and maximum allowable bandwidth are both parameters of an individual flow.

The bandwidth management and admission control mechanisms can generally be categorized as centralized or distributed. In the centralized approach, the bandwidth reservation information is maintained by a centralized bandwidth broker (BB) in each network domain. Bandwidth management and admission control modules are installed and executed only at the centralized BB [23], [24]. Although the centralized mechanism tends to achieve optimized network resource utilization, it has the singlepoint failure problem and scalability issue [3], [14]. Thus, the distributed approach is used in this work, as described in Section 1.1, where the bandwidth management and admission control modules are installed and executed at individual routers rather than at a centralized BB. The distributed approach is closer to the current Internet implementation. Nevertheless, the proposed constraint model, preemption algorithms, and admission control mechanism are feasible for the networks applying the centralized approaches.

1.1 Network Model and Operations We consider an MPLS network where DS-TE is applied. Each LSP is classified into one CT and assigned with one holding priority and one setup priority. Hence, an LSP establishment request contains four parameters (bw, ct, hp, sp), indicating that bw amount of bandwidth is requested for establishing an LSP of class type ct at holding priority hp and setup priority sp. Note that connections in the same CT can have different priorities so that they have different priorities to access and retain the resources [6]. With the distributed bandwidth management, there are four main steps in establishing a new LSP. First, the source node computes a route to the destination based on the network topology, the requested bandwidth bw, and the available bandwidth of class type ct on all the links along the path. Second, the source sends a request with parameters (bw, ct, hp, sp) to all the routers along the computed path. Third, each router on the path exercises admission control and sends a positive reply to the source if its outgoing link has enough free bandwidth available to the new connection. Fourth, if there is not enough free bandwidth, then the router would activate bandwidth preemption and return a positive reply if the preemption is successful; otherwise, it would return a negative reply. If all the routers along the path return positive replies, then the LSP setup is successful, and they would reserve the requested bandwidth on the output links. This paper addresses bandwidth management issues at the LSP level. The packet level behaviors (for example, burstiness) are contained in the bandwidth requirement of individual LSPs, which may be calculated as the effective bandwidth of an LSP on the basis of forecast or measurement. This kind of approach was used in [11], [12], [13]. 1.2 Contributions There are five major contributions in this paper. First, we conduct DS-TE bandwidth management in two dimensions: CT and priority. It allows us to use matrices in bandwidth management. Second, we propose a new constraint model in which both minimum guaranteed and maximum reservable constraints are configured for each CT. The key novel idea is a use it or lend it strategy, which guarantees a

1322

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

VOL. 18,

NO. 9,

SEPTEMBER 2007

minimum bandwidth for each CT without causing fragmentation and waste. Third, three novel bandwidth preemption algorithms are presented for the MAM, the RDM, and the proposed MaxMin constraint model, respectively. In addition to the objective of minimizing the disruption to the existing connections as in [15], [16], [17], [18], our design aims at achieving preemption fairness and performance robustness. We propose that the preemption decision be made not only on the basis of priority level, but also on the basis of the reservation status and the constraint model characteristics. This way, the bandwidth reservation of the overloaded CTs is more susceptible to preemption than that of the normal load and underloaded ones so that better robustness and fairness can be achieved. The proposed algorithms also improve bandwidth efficiency. Fourth, we present a new DS-TE admission control mechanism. Unlike the traditional methods that only consider the aggregate available link bandwidth, the proposed mechanism makes admission control decision on the basis of the incoming connections CT, its priorities, the requested bandwidth, and the available bandwidth of the CT. It is feasible for the DS-TE environment. Fifth, we conduct an extensive simulation study to demonstrate the effectiveness of the proposed bandwidth management framework. When compared with the existing schemes, the proposed constraint model and preemption algorithms improve not only bandwidth efficiency, but also robustness and fairness. They achieve significant performance improvement for the well-behaving traffic classes in terms of both bandwidth utilization and bandwidth blocking and preemption probability. We also provide guidelines for selecting different DS-TE bandwidth management mechanisms. The rest of this paper is organized as follows: In Section 2, we first review two existing DS-TE constraint models and then put forward a new model. In Section 3, three bandwidth preemption algorithms are designed for three DS-TE constraint models, respectively. The DS-TE admission control mechanism is presented in Section 4. The performance evaluation is in Section 5. Finally, Section 6 gives conclusions.

As will be seen in both Sections 3 and 4, with the matrices R and A recording the bandwidth reservation and availability, respectively, the computation complexity of bandwidth allocation, preemption, and admission control is independent of the number of admitted connections and the network size. Thus, we have a scalable bandwidth management framework. Furthermore, compared with the traditional one-dimensional schemes that only considered the classes, the complexity of our two-dimensional mechanisms is hardly affected because both the number of priorities and the number of CTs are hard-coded (for example, four CTs and eight priorities) in the algorithms, not the variables.

2.1 Maximum Allocation Model (MAM) The MAM was proposed in [7] to enforce a maximum bandwidth allocation constraint for each CT. It has the following simple rules:
The bandwidth reserved by all the connections of CT i should not exceed the bandwidth constraint BCi P of CT i. That is, NP P 1 Ri; j BCi. j0 2. The total reserved bandwidth should not exceed the PNCT 1 PNP P 1 Ri; j C, link capacity; that is, i0 j0 where C is the link capacity. 3. For improving bandwidth efficiency, the sum of the bandwidth constraints is allowed to exceed the link PNCT capacity. That is, i0 1 BCi ! C. The available bandwidth for the TE class k can be computed as follows, where the TE class k is associated with class type ct and preemption priority sp in this paper: 1. availBwTE-class k Act; sp   sp NCT 1 X X sp X Rct; j; C Ri; j : min BCct
j0 i0 j0

2.2 Russian Doll Model (RDM) The RDM was proposed in [8] to enforce different bandwidth constraints for different groups of CTs. Its rules are described as follows:
The RDM applies cascaded sharing, with each bandwidth constraint being the upper bound of the bandwidth reservation of a group of CTs; that is, PNCT 1 PNP P 1 Ri; j BCb for 0 b NCT 1. ib j0 2. 0 BCi BCj C for 0 j < i NCT 1. The available bandwidth of TE class k can be computed as follows: 1. availBwTE-class k Act; sp ( ) NCT 1 X X sp Rl; j : min BCi
0 i ct li j0

BANDWIDTH ALLOCATION

We first introduce two matrices to be used in this paper. A matrix R is used to record bandwidth reservation, with its element Ri; j being the bandwidth reserved by the connections of CT i at holding priority j. R is an NCT NP P matrix, where NCT is the number of CTs, and NP P is the number of priority levels. Every time a certain amount of bandwidth bw is granted to or released from class type ct at priority level pp, the matrix R is updated as follows: Rct; hp Rct; hp bw; 1

where bw has a positive value for a bandwidth grant and a negative value for a bandwidth release. A matrix A is used to record bandwidth availability, with each element Ai; j recording the available bandwidth to the LSPs of CT i at setup priority j. Because DS-TE supports up to eight TE classes [6], matrix A has up to eight nonzero elements, which are advertised to other network nodes for the purpose of constraint-based routing.

It is obvious that if preemption is precluded, then the RDM cannot guarantee bandwidth isolation across CTs, as the link capacity could be monopolized by CT0.

2.3

The Proposed Max-Min Bandwidth Constraint Model Both the MAM and the RDM have no minimum bandwidth guarantees for CTs; thus, they cannot ensure that the service rate of the applications with stringent QoS requirements is independent of the intensity of other traffic classes. To

SHAN AND YANG: BANDWIDTH MANAGEMENT FOR SUPPORTING DIFFERENTIATED-SERVICE-AWARE TRAFFIC ENGINEERING

1323

address this issue, we propose a new constraint model on the basis of a conventional Max-Min approach in [10], in which both the minimum guaranteed and the maximum reservable bandwidth constraints are configured for each CT. The conventional approach could cause bandwidth fragmentation and waste because the minimum bandwidth of one CT cannot be used by the connections of other CTs even if it is idle. To address this problem, we devise a novel use it or lend it strategy and an associated preemption scheme in the new model. Specifically, if a certain CT consumes less than its minimum bandwidth, its unused minimum bandwidth can be lent to other CTs connections with a nonzero holding priority. In case that it increases its demand to its minimum guaranteed bandwidth later on, it is able to preempt the connections of other CTs borrowing bandwidth so as to obtain at least its minimum bandwidth. Note that connections with zero setup or holding priority are not allowed to borrow the minimum guaranteed bandwidth of other CTs because they have the highest priority and cannot be preempted once they are established. The proposed Max-Min model has five rules: CT i has a maximum reservable bandwidth constraint PNP P 1 RBmax i, where 0 i NCT 1. That is, Ri; j RBmax i. j0 2. CT i has a minimum guaranteed bandwidth constraint GBmin i. Obviously, GBmin i RBmax i, where 0 i NCT 1. 3. The sum of the maximum reservable bandwidth is allowed to exceed the link capacity; that is, PNCT 1 RBmax i ! C. i0 4. To avoid congestion, the sum of the minimum guaranteed bandwidth should not exceed the link P capacity; that is, NCT 1 GBmin i C. i0 5. The total reservation should not exceed the link PNCT 1 PNP P 1 Ri; j C. capacity; that is, i0 j0 Thus, the available bandwidth of TE class k can be computed as follows: 1. availBwTE-class k Act; sp 8 > minff2 f3 ; RBmax ct > > < Rct; 0g for sp 0 > minff2 f3 f4 f5 ; > > P : RBmax ct sp Rct; jg for sp > 0; j0 where x max0; x shared C
NCT 1 X i0

f4

X
i6ct

GBmin i

NP P 1 X j0

! Ri; j ; 9

f5 f1 shared :

10

The available bandwidth availBwTE classk in (4) is calculated differently for zero and nonzero setup priorities. Because the zero-priority connections cannot be preempted once they are established, and they are not allowed to borrow the minimum bandwidth from other CTs. The fraction of link bandwidth shared in (5) is shareable among all the CTs. In (6), f1 computes the bandwidth reservation of the connections with priorities higher (numerically smaller) than or equal to sp, which either occupies the shared bandwidth or borrows other CTs minimum guaranteed bandwidth. In (7), f2 computes the portion of the class type cts minimum guaranteed bandwidth, which is available to the new connections of class type ct with setup priority sp. In (8), f3 is the portion of the shared bandwidth that is available to the new connections of class type ct with setup priority sp. PNP P In (9), GBmin i j0 1 Ri; j computes the fraction of the CT is minimum guaranteed bandwidth that is not reserved by CT i; hence, it can be borrowed by other CTs. Therefore, f4 computes the total amount of the minimum guaranteed bandwidth belonging to the CTs other than ct, which can be borrowed by other CTs. In (10), f5 is the reservation of the connections with priorities higher than or equal to sp, which borrows other CTs minimum bandwidth. f4 f5 is the amount of bandwidth that can be borrowed by the new connections of class type ct with nonzero sp. Although the proposed model has different configuration and sharing methods from both the MAM and the RDM, a router implementing it is able to interoperate with other nodes implementing the MAM or the RDM. The IETF does not mandate the bandwidth constraint model used at routers. That is, the nodal bandwidth management scheme can be proprietary without affecting interoperability. The only requirement is that each node maintains a record of available bandwidth per TE class, which is computed according to the constraint model in use and advertised by the Open Shortest Path First Protocol (OSPF) or Intermediate System-to-Intermediate System Protocol (IS-IS) routing system [22].

3
GBmin i; ! 5

BANDWIDTH PREEMPTION

f1

NCT 1 X i0

sp X j0

Ri; j GBmin i ! Rct; j ;

f2

GBmin ct

sp X j0

f3 shared f1 ;

When a router receives a request (bw, ct, hp, sp), it may discover that the requested bandwidth bw is less than or equal to the available bandwidth Act; sp of class type ct at priority level sp, but there is inadequate unreserved bandwidth to accommodate the new demand because a portion of the available bandwidth has been reserved by the existing connections with holding priorities lower than sp. In this case, if bandwidth preemption is allowed, then a certain amount of bandwidth would be preempted from the existing connections with lower holding priorities so that the new request can be accepted. Otherwise, the request would be rejected. First, we compute the unreserved bandwidth of the TE class k at the arrival time of the new request (bw, ct, hp, sp):

1324

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

VOL. 18,

NO. 9,

SEPTEMBER 2007

( unrsvdTE-class k min C
Npp 1 X j0

NCT 1 NP P 1 X X i0

Ri; j; BCct

j0

Rct; j

when the MAM is used; 11 ( unrsvdTE-class k min


0 h ct NCT 1 NP P 1 X X ih j0

) Ri; j

BCh

when the RDM is used; 12 unrsvdTE-class k 8  > > min CPNCT 1 PNP P 1 Ri;j; > i0 j0 > > > >  > > > > RBmax ctPNP P 1 Rct;j > > j0 > > > > <  > > min s s s s ; > 1 2 3 4 > > > > >  > > NP P 1 > > RBmax ct P Rct;j > > > j0 > > : where s1 shared
NCT 1 X i0 NP P 1 X j0

when the proposed model is used for sp > 0

13a

bandwidth. The first is the unreserved portion of shared bandwidth, which is s1 in (14). The second is the unreserved portion of GBmin ct. Equation (16) calculates s3 , which is the total amount of bandwidth borrowed by all CTs, and (17) obtains s4 , which is the amount of bandwidth that can be borrowed from CTs other than ct. Thus, s3 s4 is the bandwidth borrowed from ct. Equation (15) computes s2 , which is the portion of GBmin ct that is not reserved by the class type ct. Hence, s2 s3 s4 obtains the unreserved portion of GBmin ct. Therefore, we obtain (13b). In Sections 3.1, 3.2, and 3.3, we present three bandwidth preemption algorithms for the MAM, the RDM, and the proposed Max-Min model, respectively. They are responsible for locating both the CTs and priorities for bandwidth preemption, and sending preemption requests to the connection management. It is up to the connection management to select the connections and to decide whether to tear down the connections or to reduce the bandwidth allocation of some elastic connections, which is not the concern of this paper. In each algorithm, a matrix R 0 records the reservation after the preemption, and it is used to update R if and only if the preemption is successful. A vector b bw is used to record the amount of bandwidth to be preempted, and both vectors b ct and b pp record the CTs and the priority levels, respectively, from which b bw is to be preempted.

when the proposed model is used for sp 0;

13b !

Ri; j GBmin i ! Rct; j ! ;

14

s2

GBmin ct

NP P 1 X j0

15 !

s3

NCT 1 X i0

NP P 1 X j0

Ri; j GBmin i

shared ! Ri; j :

16

s4

X
i6ct

GBmin i

NP P 1 X j0

17

Both (11) and (12) compute the unreserved bandwidth of the TE class k when the MAM and the RDM are used, respectively. Both (13a) and (13b) compute the unreserved bandwidth when the proposed Max-Min model is used. A new connection with nonzero setup and holding priorities can be accommodated with any unreserved portion of the link P P capacity; that is, C NCT 1 NP P 1 Ri; j. Besides, the i0 j0 accommodation of the new connection should not cause class type cts reservation to exceed RBmax ct. Hence, we obtain (13a). Because the bandwidth once allocated to a zero-priority connection cannot be preempted from it, the new connection with zero priority can only be allocated with two portions of

3.1 Preemption Algorithm for MAM A simple preemption algorithm for the MAM is provided in Fig. 1. It is activated by a router when the requested bandwidth bw in an incoming request (bw, ct, hp, sp) of TE class k is smaller than or equal to availBwk, as obtained in (12), but larger than the unreserved bandwidth unrsvdk, as calculated in (11). Unlike the basic preemption rules in [4], [5], where the preemption decision is made based only on the priority level, we propose to make preemption decisions based on both the priority level and the bandwidth reservation status so as to achieve preemption fairness and to minimize the amount of the bandwidth preempted. In Fig. 1, lines 3-4, we check whether granting bw to the new connection of class type ct would cause the constraint BCct to be exceeded. If so, we need to preempt bump bw amount of bandwidth from class type ct, where bump bw is calculated in line 5. The while loop (lines 6-11) searches the row ct of matrix R 0 for the nonzero items with the lowest priorities until bump bw amount of bandwidth is located for preemption and updates the vectors b bw b ct and b pp bw, ct, pp, where b bwnb is the amount of bandwidth to be preempted from the connections of CT b ctnb at priority level b ppnb. In Fig. 1, lines 13-15, we identify whether granting bw to the new connection would cause the total reservation to exceed the link capacity C. If so, bump bw amount of bandwidth needs to be preempted, which is computed in line 14. The while loop in line 15 searches in matrix R0 for the nonzero items with the lowest priorities until bump bw amount is bw, ct, pp, located for preemption and updates b bw b ct b pp and R0 . This way, the low-priority connections are preempted before the high-priority ones, which minimizes the number of rerouted sessions. At last, a bandwidth preemption request ct, pp, with parameters b ct b pp and b bw is sent to the connection management, which selects connections for disestablishment or bandwidth compression and for returning an acknowledgment of preemption success or failure.

SHAN AND YANG: BANDWIDTH MANAGEMENT FOR SUPPORTING DIFFERENTIATED-SERVICE-AWARE TRAFFIC ENGINEERING

1325

Fig. 1. The bandwidth preemption algorithm for MAM.

3.2 Preemption Algorithm for RDM Because the RDM has a cascaded bandwidth sharing mechanism, where a certain CT i is restricted by a set of bandwidth constraints BC0; . . . ; BCi, we propose to design its preemption algorithm based on the cascaded sharing characteristics, in addition to the priority levels and bandwidth reservation status, as shown in Fig. 2. The for loop in line 3 varies the value of h. For any value of h, it is determined whether granting bandwidth bw to the class type ct would cause the constraint BCh being exceeded. The key idea is to check the constraints in the sequence of increasing constraint sizes, that is, in the order of BCct; . . . ; BC0. The purpose is to prevent preempting more bandwidth than necessary. If the largest constraint BC0 were checked first, then some low-priority connections of CT0 would get preempted unnecessarily because later checks of BCh for h 1; . . . ; ct would still require further preemption of the CT h connections. This design also provides better fairness, as BCct is the first to be checked when class type cts request is processed. If it is determined that accommodation of the requested bw would cause BCh to be exceeded, then the while loop

in line 6 searches the rows h; . . . ; NCT 1 of matrix R0 for the nonzero items with the lowest priorities until bump bw amount of bandwidth is located for preemption and bw, ct, updates the vectors b bw b ct and b pp and the matrix R0 .

Preemption Algorithm for the Proposed Max-Min Model The preemption algorithm for the proposed Max-Min model is similar to that for the MAM. The only difference lies in the while loop (Fig. 3, lines 15-23), where a search is conducted in matrix R 0 for the nonzero items with the lowest priorities until bump bw amount of bandwidth is located for preemption. The searched nonzero item R0 i; j PNP P must satisfy j0 1 R0 i; j > GBmin i for i 6 ct because a new connection of class type ct is not allowed to preempt another CT is existing connections if CT is reservation is below its guaranteed bandwidth GBmin i. Thus, the maximum amount of bandwidth that can be preempted from a PNP P CT i 6 ct is j0 1 R0 i; j GBmin i.

3.3

Fig. 2. The bandwidth preemption algorithm for RDM.

1326

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

VOL. 18,

NO. 9,

SEPTEMBER 2007

Fig. 3. Preemption algorithm for the proposed Max-Min model.

ADMISSION CONTROL

The DS-TE admission control procedure is presented in Fig. 4, which is activated by a router when a bandwidth request (bw, ct, hp, sp) is received. The admission control decision must be made on the basis of the incoming connections requested bandwidth, CT, priorities, and the available bandwidth of the CT. The algorithm returns the decision of either accepting or rejecting the new request and both the updated matrices R and A. First, it determines whether the request is for granting or releasing bandwidth. If the value of bw is negative, then it is a bandwidth releasing request, which is handled in lines 24-26. Otherwise, it is a bandwidth demand, and the requested bandwidth bw is compared with the available bandwidth Act; sp (line 3). If bw > Act; sp, then the request is rejected (line 22). Otherwise, the unreserved bandwidth unrsvdk is computed (line 4). If bw unrsvdk, then there is adequate free bandwidth available to accommodate the new demand, and the request is accepted (lines 6-8). Otherwise, the free bandwidth is inadequate, and the preemption is needed. If preemption is not allowed, then the request is rejected (line 20). Otherwise, the preemption procedure is invoked. If the preemption is successful, then the request is accepted (lines 13-16); otherwise, the request is rejected (line 18).

PERFORMANCE EVALUATION

Simulations have been conducted to evaluate the performances of the DS-TE bandwidth constraint models and preemption algorithms. In our simulations, there are three CTs, and the LSPs in each CT have two possible priorities. Priority-zero LSPs have higher priority and can preempt the priority-one LSPs. We manage bandwidth at the LSP level,

and each LSP may carry a number of service level Internet flows (for example, voice and data); thus, it is reasonable to assume that the amount of bandwidth requested by one LSP is uniformly distributed between 64 kilobits per second and 5 megabits per second. The establishment requests of LSPs arrive according to a Poisson distribution, which is a reasonable assumption, since we are dealing with LSP session arrivals and not individual packet arrivals. The holding times of LSPs are modeled with the exponential distribution. Six bandwidth management mechanisms are studied, including the basic MAM (bMAM), the basic RDM (bRDM), the enhanced MAM (eMAM), the enhanced RDM (eRDM), the proposed Max-Min, and the conventional Max-Min mechanisms. Both the bMAM and the bRDM use the basic preemption rules in [4], [5]; that is, a connection with a high setup priority is allowed to preempt an existing connection of lower holding priority. The eMAM and the eRDM use the proposed preemption algorithms in Sections 3.1 and 3.2, respectively. The proposed Max-Min mechanism uses the preemption algorithm in Section 3.3. The conventional MaxMin mechanism in [10] does not use preemption. In the simulations, the link capacity is 100 Mbps. In two MAM and two Max-Min mechanisms, the maximum bandwidth constraints of CTs 0, 1, and 2 are set, respectively, to 80 percent, 75 percent, and 55 percent of the link capacity. In two RDM mechanisms, BC0, BC1, and BC2 are set, respectively, to 100 percent, 75 percent, and 55 percent of the link capacity, because BC0 is inherently equal to 100 percent of the link capacity in the RDM. In two Max-Min mechanisms, the minimum guaranteed bandwidth constraints of CTs 0, 1, and 2 are set, respectively, to 25 percent, 30 percent, and 40 percent of the link capacity. After running simulations with different traffic loads, we obtain that bandwidth blocking and preemption rates are in the range of 104 102 when the loads of CT0, CT1, and CT2

SHAN AND YANG: BANDWIDTH MANAGEMENT FOR SUPPORTING DIFFERENTIATED-SERVICE-AWARE TRAFFIC ENGINEERING

1327

Fig. 4. Admission control mechanism.

are 5, 5, and 4 erlangs, respectively. This set of traffic loads can be viewed as the forecast loads, and satisfying the QoS can be obtained when the bandwidth constraints are configured accordingly as above. It is also called the normal load. We study two scenarios to evaluate the performances of the above six mechanisms. In both scenarios, the CT1 and CT2 have their normal loads, whereas the traffic load of CT0 increases from a lighter-than-normal load to a heavier one, which brings the total load on the link from the normal to overloaded situation. We investigate different behaviors of six mechanisms under both the normal-load and overloaded situations in Scenario 1. We study the impact of the percentage of the LSPs with priority zero (the highest priority) on the performances, where the percentage of the CT0 LSPs with priority zero increases from 50 percent in Scenario 1 to 75 percent in Scenario 2.

5.1 Scenario 1 In this scenario, the CT0 LSPs have an average holding time of 33.3 slots, and its interarrival time decreases from 33.3 to 1.33 slots; that is, its traffic load increases from 1 to 25 erlangs. The CT1 LSPs have a load of 5 erlangs, with an average interarrival time of 4 slots and an average holding time of 20 slots. The CT2 LSPs have a load of 5 erlangs, with an average interarrival time of 6.25 slots and an average holding time of 25 slots. Half of the LSPs in each CT are assigned with priority zero, whereas the other half has priority one. Note that it is for the sake of simplicity that we assign all LSPs in the same CT with the same average interarrival time and the same average holding time in our simulations. In reality, the LSPs in the same CT may carry a number of various service level flows, thus having different traffic parameters. In Fig. 5a, the bandwidth utilization of CT0 increases with the increasing load of CT0. The increase is slower when two Max-Min mechanisms are used than that when the RDM or the MAM is used. This is because a minimum guaranteed bandwidth is provisioned for each CT in the

Max-Min mechanisms, which provides better bandwidth isolation and makes the misbehaving CT0 less intrusive. In Fig. 5b, the bandwidth blocking and preemption probability of CT0 increases slightly with the increasing CT0 load when the RDM or the MAM is used, and it increases sharply when the Max-Min mechanisms are used. It means that the misbehaving CT0 is punished more severely in the Max-Min models. The eRDM and eMAM induce sharper increases than the two basic methods. This is because preemption fairness is supported in our proposed preemption algorithms, and the reservation of the misbehaving CT0 is more likely to be preempted than that of the conforming CTs. In Figs. 5c and 5e, the bandwidth utilization of CT1 and CT2 decrease with the increasing load of CT0 when bRDM, bMAM, eMAM, or eRDM is used in the order of decreasing speed. It means that the enhanced methods provide better robustness for the conforming CTs than the two basic ones. When the Max-Min mechanisms are used, the bandwidth utilization of CT1 and CT2 only decrease slightly, which means that the conforming CTs are well protected. In Figs. 5d and 5f, the bandwidth blocking and preemption probabilities of CT1 and CT2 increase slightly with the increasing CT0 traffic load when the two Max-Min mechanisms are used. They increase quickly when the RDM or the MAM is used, and the eRDM and the eMAM induce smoother increases than the two basic methods. Again, the proposed preemption algorithms achieve significant improvement over the basic preemption rules in terms of better protection and robustness for the conforming CTs. In Fig. 5g, the eRDM achieves the highest average bandwidth utilization, and the eMAM obtains the second highest utilization. The utilization achieved by the proposed Max-Min mechanism is nearly 10 percent lower than the highest value. This is because, as per (2), (3), and (4), the available bandwidth for the zero-priority LSPs is less than that for the nonzero-priority LSPs in the proposed Max-Min model. With 50 percent of LSPs having priority zero, it is an

1328

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

VOL. 18,

NO. 9,

SEPTEMBER 2007

Fig. 5. Simulation results of Scenario 1.

expected result. The eRDM and eMAM achieve two highest bandwidth efficiency values, but they induce four to seven times higher bandwidth blocking and preemption rates to the conforming CT1 and CT2 than the proposed Max-Min mechanism. Thus, there is a trade-off between bandwidth efficiency and robustness. Summarizing the results, we can see that with our proposed preemption algorithms, the eRDM and the

eMAM obtain better fairness, robustness, and efficiency than the basic methods. The proposed Max-Min mechanism achieves much better robustness and protection than both the eRDM and the eMAM, but it may not be the most desirable mechanism when there are a high percentage of connections with the highest priority (say, 50 percent) due to the low bandwidth efficiency.

SHAN AND YANG: BANDWIDTH MANAGEMENT FOR SUPPORTING DIFFERENTIATED-SERVICE-AWARE TRAFFIC ENGINEERING

1329

Fig. 6. Simulation results of Scenario 2.

5.2 Scenario 2 In Scenario 2, the average interarrival time and holding time of each CT are the same as that in Scenario 1. The only difference from Scenario 1 is that 75 percent of the CT0 LSPs have priority zero, and 25 percent of the CT0 LSPs are assigned with priority one. It means that CT0 is more misbehaving than it was in Scenario 1.

Comparing Fig. 6 with Fig. 5, we can see that when the proposed Max-Min mechanism is used, the CT0 obtains lower bandwidth utilization and higher blocking and preemption probability than that in Scenario 1. This is because the zero-priority LSPs have less available bandwidth than the nonzero-priority LSPs in the proposed mechanism. With 25 percent more CT0 LSPs having zero priority than that in Scenario 1, less bandwidth is available to CT0, which

1330

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

VOL. 18,

NO. 9,

SEPTEMBER 2007

causes lower bandwidth utilization and higher blocking rate for CT0. We also observe that both CT1 and CT2 obtain higher bandwidth utilization and lower blocking and preemption probability than that in Scenario 1. Thus, when the proposed mechanism is used, if a certain CT becomes more misbehaving in terms of more connections with the highest priority than forecast, then its performances will deteriorate, whereas other conforming CTs will obtain better performances. When the RDM or the MAM is used, the CT0 achieves slightly higher bandwidth utilization and lower blocking and preemption probability than that in Scenario 1, whereas CT1 and CT2 obtain slightly lower bandwidth utilization and higher blocking and preemption probability. This is because the available bandwidth computation is the same for the zero and nonzero-priority LSPs in the RDM and the MAM. With 25 percent more zero-priority CT0 LSPs than that in Scenario 1, more CT0 LSPs can be established by preempting lower priority LSPs of other CTs. Hence, the misbehaving CT0 obtains better performances than that in Scenario 1, whereas the CT1 and CT2 performances become worse. Thus, with the RDM or the MAM, if a certain CT becomes more misbehaving in terms of more connections with the highest priority than forecast, then its performances will improve, whereas the performances of other conforming CTs will deteriorate. In Fig. 6g, it can be seen that the total bandwidth utilization achieved by the proposed Max-Min mechanism is lower than that in Scenario 1 because 25 percent more CT0 LSPs have priority zero, which have less available bandwidth than the nonzero-priority LSPs in the proposed mechanism. The RDM and the MAM achieve higher total bandwidth utilization than that in Scenario 1, but they induce much higher blocking and preemption probabilities to the conforming CT1 and CT2, as shown in Figs. 6d and 6f. Again, there is a trade-off between the bandwidth efficiency and robustness.

and further result in low bandwidth utilization. It is better to apply traffic shaping to make the traffic flows more conforming, rather than provisioning small bandwidth constraint values. Note that one-hop simulation is sufficient in this paper because we focus on the nodal bandwidth management mechanisms, without addressing the routing aspect and network load balancing. The proposed mechanisms manage bandwidth on both CT and priority dimensions, without considering the selection of individual end-to-end connections for preemption or bandwidth compression. The multihop simulation will be useful in our future works to investigate the impact of the nodal mechanisms on the routing computation and global network performance. Our future works also include the mathematical analysis of the constraint models, the scheduling algorithms for implementing the constraint models, and so forth.

CONCLUSIONS

We have presented a bandwidth management framework for the support of DS-TE, in which the link bandwidth is managed in two dimensions: CT and priority. We put forward a Max-Min constraint model, in which we proposed a novel use it or lend it strategy. Furthermore, we designed three new bandwidth preemption algorithms for three DS-TE constraint models, respectively. An extensive simulation study demonstrated that the proposed Max-Min constraint model and preemption algorithms improve not only bandwidth efficiency, but also robustness and fairness. They achieve significant performance improvement for the wellbehaving traffic classes in terms of both bandwidth utilization and bandwidth blocking and preemption probability. We also provided guidelines for selecting different DS-TE bandwidth management mechanisms.

REFERENCES
[1] F. Le Faucheur et al., MPLS Support of Differentiated Services, IETF RFC 3270, May 2002. [2] S. Blake et al., An Architecture for Differentiated Services, IETF RFC 2475, Dec. 1998. [3] D. Awduche et al., RSVP-TE Extension to RSVP for LSP Tunnels, RFC 3209, Dec. 2001. [4] D. Awduche et al., Overview and Principles of Internet Traffic Engineering, RFC 3272, May 2002. [5] D. Awduche et al., Requirements for Traffic Engineering over MPLS, IETF RFC 2702, Sept. 1999. [6] F. Le Faucheur and W. Lai, Requirements for Support of Differentiated Service Aware MPLS Traffic Engineering, IETF RFC 3564, July 2003. [7] F. Le Faucheur et al., Maximum Allocation Bandwidth Constraints Model for Diff-Serv-Aware MPLS Traffic Engineering, IETF Internet draft, 2004. [8] F. Le Faucheur et al., Russian Dolls Bandwidth Constraints Model for Diff-Serv-Aware MPLS Traffic Engineering, IETF Internet draft, 2004. [9] W.S. Lai, Traffic Engineering for MPLS, Proc. Internet Performance and Control of Network Systems III Conf., vol. 4865, pp. 256267, July 2002. [10] F. Kamoun and L. Kleinrock, Analysis of Shared Finite Storage in a Computer Network Node Environment under General Traffic Conditions, IEEE Trans. Comm., vol. 28, no. 7, pp. 992-1003, July 1980. [11] D. Mitra and I. Ziedins, Hierarchical Virtual Partitioning: Algorithms for Virtual Private Networking, Proc. IEEE GLOBECOM 97, pp. 1784-1891, 1997. [12] S.C. Borst and D. Mitra, Virtual Partitioning for Robust Resource Sharing: Computational Techniques for Heterogeneous Traffic, IEEE J. Selected Areas in Comm., vol. 16, no. 5, pp. 668-678, June 1998.

5.3 Discussions We have also studied many other scenarios, and the results are not presented due to space limitation. In one scenario, we use the same traffic parameters for each CT as that in Scenario 1, except that only 20 percent of LSPs have priority zero. We observe that the bandwidth utilization achieved by the proposed Max-Min mechanism is much higher than that in Scenario 1 and only 1.5 percent lower than the highest value achieved by the eRDM. This is expected because having fewer zero-priority LSPs implies that more bandwidth can be shared flexibly among CTs in the proposed Max-Min mechanism. Thus, the proposed MaxMin mechanism is the most desirable mechanism when there is just a small percentage of connections with the highest priority (say, 20 percent). In another scenario, we use the same traffic parameters as those in Scenario 1. The only difference is that the maximum bandwidth constraint of CT0 is reduced to 50 Mbps. We observe that when the MAM or the proposed Max-Min methods are used, the total bandwidth utilization is much lower than that in Scenario 1, whereas the conforming CT1 and CT2 obtain better performances. Thus, reducing the bandwidth constraint values may not be an appropriate way to improve bandwidth protection because small constraint values put a curb on bandwidth sharing

SHAN AND YANG: BANDWIDTH MANAGEMENT FOR SUPPORTING DIFFERENTIATED-SERVICE-AWARE TRAFFIC ENGINEERING

1331

[13] E. Bouillet, D. Mitra, and K.G. Ramakrishnan, The Structure and Management of Service Level Agreements in Networks, IEEE J. Selected Areas in Comm., vol. 20, no. 4, pp. 691-699, May 2002. [14] W. Jia et al., Distributed Admission Control for Anycast Flows, IEEE Trans. Parallel and Distributed Systems, vol. 15, no. 8, pp. 673686, Aug. 2004. [15] M. Peyravian and A.D. Kshemkalyani, Decentralized Network Connection Preemption Algorithms, Computer Network and ISDN System, vol. 30, no. 11, pp. 1029-1043, June 1998. [16] S. Jeon, R.T. Abler, and A.E. Goulart, The Optimal Connection Preemption Algorithm in a Multi-Class Network, Proc. IEEE Intl Conf. Comm. (ICC 02), pp. 2294-2298, Apr. 2002. [17] J.C. de Oliveira et al., A New Preemption Policy for DiffServAware Traffic Engineering to Minimize Rerouting, Proc. IEEE INFOCOM 02, May 2002. [18] C. Scoglio et al., TEAM: A Traffic Engineering Automated Manager for DiffServ-Based MPLS Networks, IEEE Comm. Magazine, pp. 134-145, Oct. 2004. [19] R. Guerin, A. Orda, and D. Williams, QoS Routing Mechanisms and OSPF Extensions, Proc. GLOBECOM 97, pp. 1903-1908, Nov. 1997. [20] Q. Ma and P. Steenkiste, Supporting Dynamic Inter-Class Resource Sharing: A Multi-Class QoS Routing Algorithm, Proc. IEEE INFOCOM 99, pp. 649-660, 1999. [21] W.C. Lee, M.G. Hluchyj, and P.A. Humblet, Routing Subject to Quality-of-Service Constraints in Integrated Communication Networks, IEEE Network, vol. 9, no. 4, pp. 14-16, July-Aug. 1995. [22] F. Le Faucheur et al., Protocol Extensions for Support of Differentiated-Service-Aware MPLS Traffic Engineering, IETF Internet draft, Mar. 2004. [23] Z. Duan et al., A Core Stateless Bandwidth Broker Architecture for Scalable Support of Guaranteed Services, IEEE Trans. Parallel and Distributed Systems, vol. 15, no. 2, pp. 167181, Feb. 2004. [24] B. Teitelbaum et al., Internet2 QBone: Building a Testbed for Differentiated Services, IEEE Network, vol. 13, no. 5, pp. 8-16, Sept. 1999. [25] J. Ash, Max Allocation with Reservation Bandwidth Constraints Model for DiffServ-Aware MPLS Traffic Engineering and Performance Comparison, IETF Internet draft, Jan. 2004. [26] L. Kalampoukas, A. Verma, and K.K. Ramakrishnan, An Efficient Rate Allocation Algorithm for ATM Networks Providing MaxMin Fairness, Proc. Sixth IFIP Intl Conf. High-Performance Networking, pp. 143-154, Sept. 1995. [27] Y.T. Hou, S.S. Panwar, and H.H.-Y. Tzeng, On Generalized MaxMin Rate Allocation and Distributed Convergence Algorithm for Packet Networks, IEEE Trans. Parallel and Distributed Systems, vol. 15, no. 5, pp. 401-416, May 2004.

Tong Shan received the bachelors degree from the Nanjing University of Posts and Telecommunications, the masters degree from the China Academy of Posts and Telecommunications, and the PhD degree from the University of Ottawa, all in electrical engineering. Her research interests include traffic control, resource management, quality of service, network restoration, and security in both wired and wireless networks. She is a member of the IEEE.

Oliver W.W. Yang received the PhD degree in electrical engineering from the University of Waterloo, Ontario, Canada. He is currently a professor in the School of Information Technology and Engineering at the University of Ottawa, Ontario, Canada. He has worked for Northern Telecom Canada Ltd. and has done various consulting. He was an editor of the IEEE Communication Magazine and an associate director of OCIECE (Ottawa-Carleton Institute of Electrical and Computer Engineering). He is currently on the editorial board of the IEEE Communication Surveys & Tutorials. His research interests are in the modeling, analysis and performance evaluation of computer communication networks, their protocols, services, and interconnection architectures. The CCNR Lab, under his leadership, has been working on various projects in the switch architecture, traffic control, traffic characterization, and other traffic engineering issues in both wireless and photonic networks, of which the results can be found in many technical papers. He is a senior member of the IEEE.

. For more information on this or any other computing topic, please visit our Digital Library at www.computer.org/publications/dlib.

Vous aimerez peut-être aussi