Joint VM placement and routing for data center traffic engineering

Abstract

Abstract-Current data centers usually operate under poor utilization due to resource fragmentation. The hierarchical nature of data centers places a limit on the achievable aggregate bandwidth in the backbone. Suboptimal virtual machine placement also introduces unnecessary cross network traffic. In this paper, we solve a joint tenant (i.e., server or virtual machine) placement and route selection problem by exploiting multipath routing capability and dynamic virtual machine migration. These two complementary degrees of freedom: placement and routing, are mutually-dependent, and their joint optimization turns out to substantially improve data center efficiency. We propose (i) an offline algorithm that solves a static problem given a network snapshot, and (ii) an online solution for a dynamic environment with changing traffic. Leveraging and expanding the technique of Markov approximation, we propose an efficient online algorithm that requires a very small number of virtual machine migrations. Performance evaluation that employs the synthesized data center traffic traces, on various topologies and under a spectrum of elephant and mice workloads, demonstrates a consistent and significant improvement over the benchmark achieved by common heuristics used in today's data centers.

Dynamic Virtual Network Traffic Engineering with Energy Efficiency in Multi-location Data Center Networks

“…First, we started with the case of all demands being homogeneous for VN customers, i.e., we set h, r = 10, 1.65 (Case-H). In the next case, we assigned the demand to be uniformly chosen at random from the discrete values in the range given by h, r = [8,12] The choice of the above parameters in our study was motivated by the set of questions we posed in the Introduction. While we discuss a number of results using the above parameter values to answer these questions, we have two main postulates: Postulate-1: we postulate that when the bandwidth demand and the resources per request vary uniformly from an average value, the cost and the blocking would be higher compared to when the bandwidth demand and resources for each request is fixed; Postulate-2: we postulate that by taking three values for h, r in increasing order, the VN class with the lowest resource requirement would receive better treatment (lower blocking and cost) by the network than the other.…”

Resilience options for provisioning anycast cloud services with virtual optical networks

“…In the problem of survivable VNet embedding, [7] and [8] consider that the VNet is already designed and given, while in [9], [10], the authors build the most bandwidth efficient resilient VNet, under unicast traffic assumptions and using either single or multiple hop routing of requests in the virtual network. In proposing solutions for optimal server selection, as well as physical layer routing of anycast services for intra-and inter-DC networks, the resilience of the resulting virtual layer design is not considered by [11], [12]. It is important to note that we deal with a planning problem, jointly deciding on multiple VNets, and not an online VNet mapping that maps one VNet at a time (as in, e.g., [13]).…”

Allocating Compute and Network Resources Under Management Objectives in Large-Scale Clouds

“…The solution supports a broad range of management objectives with a single algorithm that performs adaptive placement. The solution is fully decentralized and highly Most works in the literature, for instance [19][20][21][22][23][24][25][26][27], address only a single management objective. The result in [28] supports multiple management objectives, but each objective with a different algorithm.…”

“…More recently, more attention has been devoted to the problem of online (or adaptive) placement, whereby a sequence of VDC or VM requests is processed in an iterative fashion [26,27]. Our work falls into the second category.…”