IEEE COMMUNICATIONS LETTERS, ACCEPTED FOR PUBLICATION 1 Virtual Overlay Network Scheduling Feng Gu, Chongyang Xie, Min Peng, C ¸ ic ¸ek C ¸ avdar, Samee Khan, and Nasir Ghani Abstract—As applications continue to expand, there is a growing need to provide advance reservation scheduling for higher-end virtual overlay network services. Hence this work presents an optimization formulation for this problem and studies several heuristic solutions using network simulation. Index Terms—Advance reservation, network scheduling, topol- ogy overlay, virtual network services I. I NTRODUCTION T HE need to schedule future user demands, i.e., advance reservation (AR), is becoming an integral requirement for a range of applications in grid-computing, e-science, storage backup, community networking, special event broad- casting, etc. Hence researchers have proposed a range of connection scheduling schemes for various AR service models (fixed start/duration, variable start/duration, etc) in bandwidth- provisioning and optical wavelength-routing networks, see [1]- [3]. Others have also looked at broader AR rerouting [4], survivability [5] and distributed implementation [6] strategies. However, with expanding user application scenarios, there is a growing need to extend AR schemes to support multipoint connectivity between several sites. Now many studies in virtual private network (VPN) design and overlay network provisioning [7] have looked at interconnection strategies over physical networks using optimization and heuristic methods. Notable examples here include the work in [8],[9] as well as dynamic solutions in [10], [11]. Earlier efforts have also studied virtual topology (VT) design for optical networks, see [12]. Nevertheless, the above efforts have only treated immediate requests that arrive in an “on-demand” or a-priori manner. In light of the above, there is a critical need to de- velop scheduling solutions for multipoint network overlay services, i.e., termed here as virtual overlay network schedul- ing (VONS). Indeed, there are no known studies in this area. Along these lines, this paper is organized as follows. An idealized integer linear programming (ILP) model for the VONS problem is first presented in Section II along with a heuristic solution in Section III using graph-theoretic schemes. Section IV then presents detailed simulation results, and conclusions and future work directions are presented in Section V. Manuscript received April 16, 2011. The associate editor coordinating the review of this letter and approving it for publication was J. Wang. F. Gu, C. Xie, and N. Ghani are with the University of New Mexico (e-mail: nghani@ece.unm.edu). M. Peng is with Wuhan University. C ¸.C ¸avdar is with KTH Royal Institute of Technology. S. Khan is with North Dakota State University. Digital Object Identifier 10.1109/LCOMM.2011.11.110819 II. OPTIMIZATION FORMULATION An ILP model of the VONS problem is presented. The formulation assumes idealized settings where all overlay re- quests are known a-priori. Consider the requisite notation. The network is modeled as a graph, (,), where is the set of router/switch nodes and the set of physical links. Without loss of generality, all links have capacity, , and connectivity is bidirectional, i.e., two uni-directional links between adjacent nodes. Each physical link ∈ also has a bandwidth-time function, i.e., (), which tracks the used capacity at future time. In addition, the -th virtual overlay network request is denoted by the 5-tuple =( , , , , ), where is the set of node sites ( ⊆ ), is the set of virtual links between nodes in , is the start time, is the stop time, and is the requested bandwidth, ≤ . Normally, request setup entails scheduling the set of connections corresponding to the virtual links in , i.e., connection endpoints designated by virtual link endpoints. As per ILP requirements, time is discretized into fixed timeslots, i.e., and are integral multiples of a fixed timeslot interval . In addition, some other variables are also defined: ∙ is the set of all VONS requests ∙ = max ∈ { } is the maximum stop timeslot across all requests ∈ ∙ ∈ is the -th node selected in ∙ , is the virtual link between and , ∕= ∙ ,, , is a binary flag which denotes virtual link occupa- tion in time slot , i.e., ,, , =0 if , does not use link ∈ at time slot ; ,, , =1 if , uses link ∈ at time slot ∙ → if ∈ is the egress node of link ∈ ; → if ∈ is the ingress node of link ∈ Using the above, the objective function is defined as: ∑ ∈ ∑ ∈ ∑ ∈ ∑ ∈ ∑ 0≤≤ ,, , (Eq.1) subject to the following constraints: ∑ → ,, , =1 ∈ , ≤ ≤ , ∈ , ∈ (Eq.2) ∑ → ,, , =0 ∈ , ≤ ≤ , ∈ , ∈ (Eq.3) ∑ → ,, , =1 ∈ , ≤ ≤ , ∈ , ∈ (Eq.4) 1089-7798/11$25.00 c ⃝ 2011 IEEE