The Application of Bandwidth Optimization ... - Semantic Scholar

3 downloads 0 Views 450KB Size Report
on the link for traffic caused by contracting of SLAs (Service. Level Agreement). It could be an important part of admission control in the SLA negotiation process ...
The Application of Bandwidth Optimization Technique in SLA Negotiation Process Srećko Krile University of Dubrovnik Department of Electrical Engineering and Computing Cira Carica 4, 20000 Dubrovnik, Croatia Tel. +385 20 445-739, Fax. +385 20 435-590, e-mail. [email protected]

Đuro Kužumilović University of Dubrovnik Department of Electrical Engineering and Computing Cira Carica 4, 20000 Dubrovnik, Croatia Tel. +385 20 445-739, Fax. +385 20 435-590, e-mail. [email protected] Abstract - In this paper we propose an efficient resource management technique for MPLS-based DiffServ networks. The service provider (SP) can predict sufficient bandwidth resources on the link for traffic caused by contracting of SLAs (Service Level Agreement). It could be an important part of admission control in the SLA negotiation process (SLA-creation). Decisions of optimal capacity arrangements from network provider (NP) have to be done interactively and dynamically. It can significantly improve QoS end-to-end routing in the moment of service invocation. The problem of bandwidth optimization can be seen as an expansion problem of link capacities. Such explicit traffic engineering technique provides the possibility to intelligently tailor the route foe each SLA traffic flow such that different parts of the network remain equally loaded. can be done dynamically but such resource management approach has to be the result of all previous contracted SLAs, that are participating with definite bandwidth in the same time period. Such bandwidth optimization technique helps to avoid the creation of bottleneck links on the path and maintains high network resource utilization efficiency. The important difference from existing traffic routing techniques is that optimal path for SLA need not to be the shortest path solution as we have in e.g. OSPF protocol (often used for traffic routing in the moment of service invocation). Index Terms - dynamic bandwidth, admission control in DiffServ/MPLS networks, SLA creation, end-to-end QoS routing, traffic routing of aggregate flows.

I. INTRODUCTION In DiffServ networks the classification of the aggregated flows is performed according to the SLA (Service Level Agreement) signed between a customer and the service provider (SP) operator. Each SLA contract specifies a time period for service utilization of defined service class and appropriate bandwidth (speed how traffic may be sent). In general, many elements define service class or QoS level. To obtain quantitative end-to-end guarantees in DiffServ/MPLS architecture, based on traffic handling mechanisms with aggregate flows, some kind of bandwidth optimization technique (on the link) has to be introduced. Multi-protocol label switching

(MPLS) has gained popularity as a technology for managing network resources and providing of performance guarantees. In this paper we are looking for optimal path provisioning for traffic flows caused by all existing SLAs participating in the same time period (former contracted SLAs); see fig. 1 and fig. 2. The main condition is: the sufficient network resources must be available at any moment of that period. In the worse case it must be sufficient for the first service class (full satisfied). Any virtual network (VN) is a subset of physical network resources purchased by an SP from the IP transport network provider (NP) to build a logical service delivery infrastructure. It can be made dynamically, using some TE technique. If we can predict sufficient link resources (without shortages) in the moment of SLA creation, the possibility of link congestions in VN will be significantly reduced in the moment of service invocation. The problems of dynamic bandwidth optimization technique are investigating. Such problem can be seen as the link capacity expansion problem (CEP) from the common source with expansion values in allowed limits. Explanation of the mathematical model for link bandwidth QoS service clasess

SLA_1 SLA_new

guaranteed (fully satisfied)

adaptive (almost satisfied) best-effort

SLA_2 SLA_3 SLA_4 SLA_5 critical period

time

Figure 1. An example of number of SLAs that participate in the same period of time; see figure 2. The bandwidth optimization and the congestion control has to be done for that critical period (duration of new SLA).

SLA_3 ingress SLA_6 egress Edge router SLA_ 3

SLA_new ingress SLA_1 ingress SLA_2 ingress SLA_4 egress

2

SLA_new egress SLA_2 egress SLA_4 ingress SLA_5 egress SLA_6 ingress

4 SLA_new

SLA_ 4

SLA_ 2

s

SLA_new

Edge router SLA_1

1

6

SLA_new

SLA_new

d

SLA_ 6

Edge router

SLA_new

3

5

Sla

core network SLA_ 5

SLA_1 egress SLA_3 egress SLA_5 ingress

Edge router

Figure 2. An example of number of SLAs in context of new SLA creation. In service specifications (details about service activation and duration) they share bandwidth on the link in the same period of time; see fig. 1.

expansion technique is given in section II. Heuristic algorithm is described in section III. Numerical examples, testing results and algorithm application are discussed in section IV. The service provider (e.g. ISP) wants to accept new SLA (traffic flow) that generates the traffic flow between edge routers. In DiffServ/MPLS architecture we call them LER (Label Edge Router). Traffic demand on exit of LER represents the sum of all ingress and egress flows. Interior routers in core network, capable to forward traffic in equivalent classes (FEC), are called LSR (Label Switching Routers). The network example that is shown in figure 1 can be a good representation of such architecture. Such aggregated flow is coming to LSR and has to be routed to destination (egress router). Packets of the same FEC are assigned the same label and generally traverse trough the same path across the MPLS network. A FEC may consist of packets that have common ingress and egress nodes, or the same service class and same ingress/egress nodes or any other combination. In this case, the FEC of appropriate service class aggregates traffic demands (new SLA and former contracted SLAs) of the same QoS level. A path traversed by an FEC is called a label switching path (LSP). In that sense network operator has to find the optimal path for the FEC without any congestion in the network. Possibility of congestion on some link (with limited bandwidth) exists, specially for definite period of time; see fig. 1. For each communication link in the DiffServ network given traffic demands can be satisfied on many (usually three) different QoS levels (with appropriate bandwidth). At the SLA level, the transport network

resources are shared by set of per-class, per ingress/egress-pair SLAs. Traffic can be satisfied on appropriate QoS level or higher, but not on lower QoS level. In fact, different LSP is used for different service class. But SP doesn’t want that transport network be divided into multiple virtual networks, one per service class. The end effect is that SP can give to premium traffic more resources, but exactly that is necessary, no less no more. In the same time SP strongly needs the optimal utilization of limited capacity resources with minimal bandwidth resources arrangement from NP. The optimal resource management problem can be seen as the link capacity expansion problem (CEP) in allowed limits; see [4] and [6]. With congestion control/expansion algorithm we can predict link resources to satisfy traffic demands caused by all accepted SLAs. If the optimal routing sequence has any link expansion with value that exceeds allowed limits, it means that link capacity on the path cannot be sufficient for such traffic. It means that new SLA cannot be accepted and must be redefined through negotiation process. For example, the customer can decide to take the adaptive QoS service class instead of fully satisfied (guaranteed) service class. Some important papers about that problem are [1], [2] and [3]. In paper [1] such bandwidth optimization technique is a part of TEQUILA architecture based on autonomic computing. In the paper [4] such algorithm is a part of service management architecture. In papers [8] and [9] similar network engineering technique is related on path provisioning.

Consider a network G (A, E) where each link is characterized by z-dimensional link weight vector, consisting of z-nonnegative QoS weights. The number of QoS measures (e.g. bandwidth, delay) is denoted by z. Weights are limited with appropriate constraints for each measure. QoS measures can be roughly classified into additive (e.g. delay) and non-additive (e.g. available bandwidth). In case of an additive measure, the QoS value of the path is equal to the sum of the corresponding weights of the link along that path. For a non-additive measure, as we have in our problem, the QoS value of the path is the minimum (or maximum) link weight along the path. In this mathematical model it is assumed that the network state information (e.g. traffic demands, link capacities, delay limits) are temporary static. In general we have multi-constrained problem (MCP) but in this paper we talk about one-dimensional link weight vectors for M+1 links on the path {wi,m, m ∈ E, i = 1, …, N} with only one constraint denoted with L. Definition of the single-constrained problem is to find a path P from node s to node d such that:

II. THE MATHEMATICAL MODEL The problem explained above can be solved as the capacity expansion problem (CEP) without shortages. Partially expansions for each link are made from common source in given limits. Transmission link capacities (bandwidth) on the path between routers are capable to serve traffic demands for N different QoS levels (service class) for i = 1,2, ..., N. Fig. 3 gives an example of network flow representation for multiple QoS levels (N) and M internal (core) routers included in the path. Let G (A, E) denote a network topology, where A is the set of nodes and E the set of links. The source and destination nodes (e.g. edge routers in IP-domain) are denoted by s and d respectively. Between them there are M interior routers on the path; see fig 3. For M interior routers on the path we have: M – 1 links + 2 = M + 1 links {m = 1, …, M+1}. In the mathematical model the following notation is used: i, j and k = QoS level. We differentiate n service classes (QoS levels).The N levels are ranked from i = 1,2,..., N, from higher to lover. m = the order number of the link on the path, connecting two successive LSR routers, m = 1, …., M+1. u,v = the order number of capacity points in the subproblem, 1 ≤ u, ..., v ≤ M+1. ri,m = traffic demand increment for additional capacity. Ii,m = the relative amount of idle capacity Ii1 = 0, Ii,M+1 = 0. xi,m = the amount of used capacity for facility. Li,m = upper bound for link capacity yi,j,m = the amount of capacity for facility i on the link m, redirected to satisfy the traffic of lower level j.

M +1 N

∑∑w

w ( P ) = min

m =1 i =1

i,m

( I i,m )

(1)

where: Ii,m ≤ L i,m

(2)

for i = 1, … , N ; m = 1, …, M+1 A path obeying the above condition is said to be feasible. Note that there may be multiple feasible paths between s and d. Capacity situation Ii,m of each link depends of expansion solutions (xi,m, yi,j,m,). Bandwidth constraints for link capacity values on the link m and for appropriate service class are denoted with Li,m (L1,m L2,m, … LN,m). Such problem is special case of the MCP

O R

1 ,N

X 1 ,1

I 1 ,1 = 0

X 1 ,2

X 1 ,M

I 1 ,3

I 1 ,2 1 ,1

1 ,2

r 1 ,1

I 2 ,1 = 0

r 1 ,M y 1 2 ,M

I 2 ,3

2 ,1

. . .

I N ,1 = 0

. . .

2 ,2 y 1 N ,1

r 2 ,2

y 2 N ,1

. . .

I 2 ,M

I 2 ,M + 1 = 0 2 ,M

y 1 N ,2

r 2 ,M

r N ,1

. . .

r N ,2

lin k 1

I N ,M

I N ,M + 1 = 0 N ,M

r N ,M

lin k 2 in t e r io r ro u te r 1

y 1 N ,M

2 N ,M

I N ,3 N ,2

. . y.

y 2 N ,2

I N ,2 N ,1

edge ro u te r

I 1 ,M + 1 = 0 1 ,M

y 1 2 ,2

I 2 ,2 r 2 ,1

I 1 ,M

r 1 ,2 y 1 2 ,1

Q oS le v e l

. . .

. . .

. . .

lin k M + 1

in t e r io r ro u te r M

edge ro u te r

Figure 3. A network flow representation of the CEP model applied for congestion control in SLA creation process. Service classes are differentiated but possible traffic conversion is possible only in direction toward higher QoS level.

problem, known as Restricted Shortest Path (RSP) problem. We can say that each weight wi,m depends of capacity situation on the link between two routers (capacity expansion or conversion, inventory or shortage etc.). That weight usually corresponds to the link expansion cost. It means that the link weight (cost) is the function of used capacity: lower amount of used capacity (smaller bandwidth with delay in acceptable limits) gives lower weight (cost). The main condition is that given traffic demands must be fully satisfied without shortages. Nonlinear cost function is necessary if link weights are not positively correlated.

III.

ALGORITHM DEVELOPMENT

The problem of the optimal bandwidth on the link for given traffic demands with different QoS service classes can be seen as the Minimum Cost Multi-Commodity Flow Problem (MCMCF) in the single (common) source multiple destination network. It is very complex problem (N-complete). Instead of nonlinear convex optimization we used two-stage network optimization technique that is more appropriate for such non-linear problem. Looking for objective function (5) on the first stage of network optimization we are calculating the minimal link weights between all pairs of capacity points. Calculation of each weight value we call it: sub-problem. Algorithm is looking for the best expansion solution among all possible (minimal cost). On the second stage of optimization we are looking for the shortest path in the network with former calculated link weights between node pairs; see fig. 4. Link capacity that is capable to serve traffic demands of service class i we call facility. It is used primarily to serve demands for QoS level i, but it can be used to satisfy traffic demands for QoS level j (j > i). Rerouting of traffic demands towards higher QoS level is the same thing as facility conversion to lower QoS level. In this model conversion of traffic demand is permitted only in the direction toward higher QoS level. It means that almost satisfied service class (adaptive or best effort) can be treated as fully satisfied (guaranteed) service class, but not vice versa. Explanation of diagram in fig. 3.: the m-th row of nodes represents a possible link capacity state of each transmission link between routers for i-th QoS level. Generalizing the concept of the capacity states for each quality level of transmission link m in which the capacity states of each link are known within defined limits we define a capacity point - αm.

αm = (I1,m, I2,m, ... , IN,m)

(3)

α1 = αM+1 = (0, 0, ... , 0)

(4)

In formulation (3) αm denotes the vector of capacities Ii,m for all QoS levels (facility types) on link m, and we call it capacity point. Each column on the flow diagrams (fig. 3.) represents a capacity point, consisting of N

capacity state values. Formulation (4) implies that idle capacities or capacity shortages are not allowed on the link between edge and interior router. Total capacity value on the link can be positive only and shortages are not allowed. Horizontal links between them represent the traffic flow between routers. Common node “O” is the source for used capacity (expansions), introducing the new traffic on the link. Vertical links represent facility conversion that is equal to rerouting of traffic demands to higher quality level (in opposite direction). If the link expansion cost corresponds to the weight of used capacity, the objective is to find optimal routing policy that minimizes the total cost incurred over the whole path between edge routers (M interior routers and M + 1 transmission links) and to satisfy given traffic demands. The flow theory enables separation of these extreme flows, which can be a part of an optimal expansion solution, from those which cannot be. With such heuristic approach we can obtain the near-optimal result with significant computational savings; see [4]. The objective function for CEP problem can be formulated as follows:

⎛ M +1 ⎧ N ⎫⎞ min⎜⎜ ∑ ⎨∑ ci ,m ⎛⎜⎝ xi ,m ⎞⎟⎠ + hi ,m (I i ,m +1 ) + g i , j ,m ( yi , j ,m )⎬ ⎟⎟ (5) ⎭⎠ ⎝ m =1 ⎩ i =1 so that we have:

I i ,m +1 = I i ,m + xi ,m −

N

∑y

j =i +1

i , j ,m

− ri ,m

(6)

I i ,m = I i ,M +1 = 0

(7)

for m = 1, 2, ..., M+1; i = 1, 2, ... , N; j = i + 1, ... , N. In the objective function the total cost (weight) on the path from edge to edge router includes some costs: the cost for capacity expansion ci,m (xi,m), the idle capacity cost

d22(1,1)

2,1 d11(1,1)

1,1

d33(1,1)

d23(1,1)

.

4,1

d12(1,1)

. . . .

d13(1,1)

d4M(1,1)

.

d2M(1,1) d2M(C2,1)

d1M(1,1)

d13(1,C4)

d11(1,C2)

3,1

.

d12(1,C3)

d23(1,C4) d2M(C2,1)

.

. ... . .

d3M(1,1)

M+1,1

d3M(C3,1) d4M(C4,1)

d22(1,C3)

2,C2 d22(C2,C3)

3,C3

4,C4 d33(C3,C4)

Figure 4. Suppose that all links in the network are known, the optimal solution for CEP can be found by searching for the optimal sequence of capacity points and their associated link values.

Di x 1 ,v x 1 ,u + 1

x 1 ,u

G u a ra n te d ( f u lly s a t is f ie d )

I 1 ,u

I 1 ,u + 1

I 1 ,u + 2

1 ,u r 1 ,u

Q oS s e r v ic e c la s s e s

r 1 ,u + 1

y 1 2 ,u

a d a p t iv e ( a lm o s t s a t is f ie d )

r 1 ,v + 1

y 1 2 ,v

I 2 ,u + 2

I 2 ,u + 1 2 ,u

2 ,u + 1 y 1 3 ,u

r 2 ,u + 1

y 2 3 ,u

I 2 ,v + 1

. .

2 ,v r 2 ,v + 1

y 1 3 ,u + 1 y 2 3 ,u + 1

I 3 ,u b e s t-e ffo rt

1 ,v

y 1 2 ,u + 1

I 2 ,u r 2 ,u

I 1 ,v + 1

. .

1 ,u + 1

I 3 ,u + 1

I 3 ,u + 2

3 ,u

I 3 ,v + 1

. .

3 ,u + 1

r 3 ,u

y 1 3 ,v y 2 3 ,v

r 3 ,u + 1

3 ,v

r 3 ,v + 1

lin k s

Figure 5. A network flow representation of a sub-problem for N=3

hi,m (Ii,m+1) as penalty cost to force the usage of the minimum link capacity (prevention of unused/idle capacity). For the link expansion in allowed limits we can set the expansion cost to zero or we can introduce different cost to each service class. Also we can introduce facility conversion cost gi,j,m (yi,j,m) that can control non-effective usage of link capacity. Costs are often represented by the fix-charge cost or with constant value. We assume that all cost functions are concave and nondecreasing, reflecting economies of scale, and they change for appropriate link. With different cost parameters we can influence to the optimization process, looking for the most appropriate routing solution.

A. Capacity Expansion Sub-problem (CES) Let the value du,v(αu, αv+1) represents the minimum cost between two capacity points. Calculation of that value is denoted as CES (Capacity Expansion Subproblem). The expansion sub-problem for N facilities i = 1, 2, .. , N on the path between routers u and v is as follows (see fig. 5.): N ⎧⎪ v ⎛ N ⎞⎫⎪ min ⎨∑ ⎜⎜ ∑ ci ,m ( xi ,m ) + hi ,m ( I i ,m +1 ) + ∑ g i , j ,m ( yi , j ,m ) ⎟⎟⎬ (8) ⎪⎩m =u ⎝ i =1 j = m +1 ⎠⎪⎭ where I i ,v +1 = I i ,u + Di (u , v ) − Ri (u , v ) (9) v

Ri (u, v) = ∑ ri ,m

Di (u,v)=



(12)

m =1

In the CEP we have to find many cost values du,v(αu, αv+1) that emanate two capacity points, from each node (u, αu) to node (v+1, αv+1) for v ≥ u; see fig. 4. The total

number of all connections between capacity points is: M ⎡ M +1 ⎤ N d = ∑ Ci ⋅ ⎢ ∑ C j ⎥ i =1 ⎣ j =i+1 ⎦

(13)

The most of the computational effort is spent on computing the sub-problem values. The number of all possible du,v values depends on the total number of capacity points. It is very important to reduce that number S x1= D1 I1,m ≤ 0

I1,m+1

D1 ≥ 0 r1,m

x3 = 1 ; (D2 + D3) I2,m = 1

(10)

I2,m+1 = 0

D2 = -2

r2,m = -1

m =u

v ⎛ ⎜ ⎜ ⎜ m =u ⎝

M +1

C p = ∑ Cm

⎞ ⎟ i , j , m ⎟⎟ j =1 ⎠ N

xi , m − ∑ y

y23 = 2 ; (- D2 )

; i≠ j

(11)

I3,m = 0

for m = 1, 2, ... , M+1; i = 1, 2, ... , N; j = i + 1, ... , N. Let Cm be the number of capacity point values at router position m (for link between core routers). Only one capacity point for the link that connects to the edge router: C1 = CM+1 = 1. The total number of capacity points is:

I3,m+1 = 0

D3 = 3 r3,m= 3

Internal router

m

m+1

Figure 6. An expansion solution

(Cp) and that can be done through imposing of appropriate capacity bounds or by introduction of adding constraints. Lot of expansion solutions is not acceptable and they are not part of the optimal sequence. Any of them if it cannot be a part of the optimal sequence is set to infinity. This is the key of the heuristic approach. The approach described in [7] requires solving repeatedly a certain single location expansion problem (SLEP). Many different expansion and conversion solutions (rerouting) can be derived, depending on Di polarity. An example of the expansion solution can be seen in fig 6. In that case premium class (fully satisfied) has to be expanded independently. Traffic demands of lowest service class can be satisfied with capacity of higher class (conversion) because traffic demands of second class has negative polarity and idle capacity (surplus) on the link is present. With such sharing approach of link bandwidth among different quality levels we can solve traffic demands efficiently.

B. Algorithm complexity Suppose that all links are known, the optimal solution for CEP can be found by searching for the optimal sequence of capacity points and their associated link state values for each router location. On that network optimization level problem can be seen as a shortest path problem for an acyclic network in which the nodes represent all possible values of capacity points and links represent sub-problem values; see fig. 4. Than Dijkstra’s algorithm or any similar algorithm can be applied. It has to be noted that the shortest path in such network need not to be the shortest path solution for routing of traffic flow (new SLA) between edge routers. Also, on this level of algorithm calculation it is very easy to introduce delay limits on the path. The number of all possible du,v values depends on the total number of capacity points. It is very important to reduce that number (Cp) and that can be done through imposing of appropriate capacity bounds or by introduction of adding constraints. The required effort for one sub-problem is O(N2M). The number of all possible du,v values depends on the total number of capacity points. If there are no limitations on capacity state (Ii,m) and expansion amount (xi,m) the complexity of such heuristic approach is pretty large and increases exponentially with N. Problem requires the computation effort of O(M3N4Ri2(N-1)). In real application we normally apply definite granularity of capacity values. It reduces the number of the capacity points significantly.

IV. TESTING RESULTS We tested our algorithm on many numerical testexamples, looking for the best routing sequence on the

Figure 7. In this numerical test-example link capacities are sufficient to satisfy traffic demands (SLAs).

Figure 8. In this numerical test-example the lack of capacity is obvious and new SLA cannot be accepted.

path. Between edge routers there are maximum six LSR (core routers) and they are connected with seven links. Traffic demands (SLAs) are overlapping in time. The heuristic algorithm in all test-examples can achieve nearoptimal expansion sequence with minimal total cost, that is equal or very close to that one we can get with algorithm based on exact approach (calculating all possible capacity points and all sub-problem values). If the expansions of link capacities are possible (in allowed limits) or the capacity surplus is obvious, the traffic demands can be satisfied. In example from figure 7. we can see that traffic demands can be fully satisfied and capacity surplus for all links are positive or zero. It means that there are no negative values; see lower diagram in fig 7. In example from figure 8. we can notice the lack of capacity on the second link (for second QoS level). It means that traffic demands cannot be fully satisfied although the surplus is obvious on third (lower) QoS level. There is no free capacity on the first QoS level (for the premium service class) to be used instead. It means that adding capacity (from NP) cannot be arranged. In that case new SLA cannot be accepted or must be redefined through negotiation process. With such management tool we can predict sufficient bandwidth on the link and possible congestions on the path very efficiently. Also, we can introduce new link capacity resources (bandwidth) in optimal way. Optimal bandwidth on the link and bandwidth arrangement for given SLAs can be done from NP dynamically. Such explicit traffic engineering technique provides the possibility to intelligently tailor the route foe each SLA traffic flow such that different parts of the network remain equally loaded. Also, it helps to avoid the creation of bottleneck links on the path and ensures optimal network resource utilization.

cost for used bandwidth and to higher resource usage efficiency. It can prevent critical network resources, not to be exhausted early and becoming a bottleneck for the network in the moment of service invocation. It means that such heuristic approach can be successfully applied for admission control in the SLA negotiation process, that is in firm correlation with resource reservation mechanisms and admission control process in DiffServ networks. The important difference from existing traffic routing techniques is that optimal path for SLA need not to be the shortest path solution as we have in e.g. OSPF protocol (often used for traffic routing in the moment of service invocation).

REFERENCES [1] Yu Cheng, R. Farha, A. Tizghadam and all, “Virtual Network Approach to Scalable IP Service Deployment and Efficient Resorce Management”, IEEE Communicattion Mag., 2005, Vol. 43, No. 10, pp.76-84. [2] K. Gopalan, Tzi-cker Chiueh, Yow-Jian Lin, “Load Balancing Routing with Bandwidth-Delay Guarantees”, IEEE Communicattion Mag., 2004, Vol. 42, No. 6, pp.108113. [3] S. Giordano, S. Salsano, G. Ventre, “Advanced QoS Provisioning in IP Networks”: The European Premium IP Projects, IEEE Communications Mag., 2003, Vol. 41 No. 1, pp.30-36. [4] S. Krile, “New SLA Creation and Optimal Resource Management”, Proc. of 3rd Conf. CSNDSP (Communication Systems, Networks and Digital Signal Processing), New Castle (UK), 2004., pp.152-155. [5] D. Kagklis, C. Tsakiris, N. Liampotis, , “Quality of Service: A Mechanism for explicit activation of IP Services Based on RSVP”, Journal of Electrical Engineering, Vol. 54, No. 9-10, Bratislava, 2003, pp.250-254. [6] S. Krile, S. Saric, “SLA Creation in Relation to

V. CONCLUSIONS In the process of SLA management in DiffServ/MPLS networks service provider (SP) needs bandwidth optimization technique to ensure sufficient resources from NP to satisfy traffic demands caused by former contracted SLAs. In this paper we proposed new bandwidth optimization technique that can ensure sufficient bandwidth on the link. It could be made dynamically through SLA negotiation process. With such tool we can significantly reduce the possibility of traffic congestions in the moment of service invocation. Proposed heuristic algorithm is based on mathematical model for the capacity expansion problem (CEP). The first aspect of algorithm is that load balancing leads to minimal

Resource Management”, Journal of Electrical Engineering (JEE), 2003., Vol. 54, No. 9-10, pp.225230. [7] S. Krile, M. Kos, “A Heuristic Approach for Path Provisioning in Diff-Serv Networks”, Proc. 7th ISSSTA (Int. Symp. On Spread-Spectrum Tech. & Application) - IEEE, Prag, 2002, pp.692-696. [8] K. Biswas, S. Ganguly and R. Izmailov, “Path provisioning for service level agreements in differentiated services networks, ICC 2002 - IEEE International Conference on Communications, vol. 25, No. 1, 2002, pp.1063–1068. [9] E. Bouillet, D. Mitra, G. K. Ramakrishnan, “The structure and management of service level agreements in networks”, IEEE Journal on Selected Areas in Communications, Vol. 20, No. 4, 2002, pp.691-699.