Today’s businesses increasingly rely on cloud computing, which brings both great opportunities and challenges. One of the critical challenges is resiliency: disruptions due to failures (either accidental or because of disasters or attacks) may entail significant revenue losses (e.g., US$ 25.5 billion in 2010 for North America). Such failures may originate at any of the major components in a cloud architecture (and propagate to others): 1) the servers hosting the application; 2) the network interconnecting them (on different scales, inside a data center, up to wide-area connections); or 3) the application itself. We comprehensively survey a large body of work focusing on resilience of cloud computing, in each (or a combination) of the server, network, and application components. First, we present the cloud computing architecture and its key concepts. We highlight both the infrastructure (servers, network) and application components. A key concept is virtualization of infrastructure (i.e., partitioning into logically separate units), and thus we detail the components in both physical and virtual layers. Before moving to the detailed resilience aspects, we provide a qualitative overview of the types of failures that may occur (from the perspective of the layered cloud architecture), and their consequences. The second major part of the paper introduces and categorizes a large number of techniques for cloud computing infrastructure resiliency. This ranges from designing and operating the facilities, servers, networks, to their integration and virtualization (e.g., also including resilience of the middleware infrastructure). The third part focuses on resilience in application design and development. We study how applications are designed, installed, and replicated to survive multiple physical failure scenarios as well as disaster failures.

/pdf/a-survey-on-resiliency-techniques-in-cloud-computing-2hn3780zwf.pdf

A Survey on Resiliency Techniques in Cloud Computing Infrastructures and Applications

Optical networking: past, present and future

In this paper, we analyze distance-adaptive transmission in cloud-ready elastic optical networks (EONs). To this end, we focus on the routing, modulation, and spectrum allocation problem, which we formulate as an integer linear programming problem, taking into account such optimization criteria (network performance metrics) as network capital expenditure and operating expenditure network cost, power consumption, and spectrum usage. The experiments are carried out with two representative network topologies and realistic data traffic models built on Cisco traffic predictions. The aim of this study is threefold. First, we examine the impact of distance-adaptive modulation formats on network performance. Second, we analyze how the optimization of a particular criterion influences other performance metrics. Finally, wefocuson the benefitsof using anycasting instead of typical unicast transmission in EONs. The results show that the use of multiple modulation formats instead of one selected format can bring significant savings in terms of all performance metrics (up to 40% for spectrum usage).

Distance-adaptive transmission in cloud-ready elastic optical networks

In this paper, we focus on the survivability of elastic optical networks (EONs) that jointly support two types of traffic demands: unicast and anycast. To provide network survivability, we apply multipath routing; i.e., we allow the splitting of a demand into a number of routing paths if the paths' combination guarantees the realization of a specific demand volume in the case of a single link failure. We formulate the corresponding optimization problem as an integer linear program (ILP) and propose a survivable multipath allocation (SMA) algorithm to solve the problem in a reasonable amount of time. Next, we perform numerical experiments to compare the efficiency (ability to provide a good-quality solution in a reasonable amount of time) of the ILP model and SMA as well as to evaluate the impact of survivable multipath routing on the objective defined as a maximum spectrum usage in EONs. Our results show that the SMA method finds good-quality solutions in a reasonable amount of time and that survivable multipath routing in EONs requires additional spectrum resources, up to 45%. However, the amount of additional resources depends on the required protection level, amount of anycast traffic, the maximum number of paths used for demand realization, and the considered network topology.

Survivable multipath routing of anycast and unicast traffic in elastic optical networks

Cloud-computing services are provided to consumers through a network of servers and network equipment. Cloud-network (CN) providers virtualize resources [e.g., virtual machine (VM) and virtual network (VN)] for efficient and secure resource allocation. Disasters are one of the worst threats for CNs as they can cause massive disruptions and CN disconnection. A disaster may also induce post-disaster correlated, cascading failures which can disconnect more CNs. Survivable virtual-network embedding (SVNE) approaches have been studied to protect VNs against single physical-link/-node and dual physical-link failures in communication infrastructure, but massive disruptions due to a disaster and their consequences can make SVNE approaches insufficient to guarantee cloud-computing survivability. In this work, we study the problem of survivable CN mapping from disaster. We consider risk assessment, VM backup location, and post-disaster survivability to reduce the risk of failure and probability of CN disconnection and the penalty paid by operators due to loss of capacity. We formulate the proposed approach as an integer linear program and study two scenarios: a natural disaster, e.g., earthquake and a human-made disaster, e.g., weapons-of-mass-destruction attack. Our illustrative examples show that our approach reduces the risk of CN disconnection and penalty up to 90 % compared with a baseline CN mapping approach and increases the CN survivability up to 100 % in both scenarios.

/pdf/disaster-survivable-cloud-network-mapping-4cawppxfs8.pdf

Disaster-survivable cloud-network mapping

Optical networks are crucial to support increasingly demanding cloud services. Delivering the requested quality of service is key to successfully provisioning end-to-end services in clouds. Therefore, as for traditional optical network services, it is of utter importance to guarantee that clouds are resilient to any failure of either network infrastructure or data centers. A crucial concept in establishing cloud services is that of network virtualization: the physical infrastructure is logically partitioned in separate virtual networks. Also, combined control of the network and data center (IT) resources is exploited. To guarantee end-to-end resilience for cloud services in such a set-up, we need to simultaneously route the services and map the virtual network, while ensuring that an alternate routing is always available. Note that the anycast routing concept applies: assigning server resources requested by the customer to a particular (physical) data center can be done transparently. This paper investigates the design of scalable optimization models to perform the virtual network mapping resiliently (for single bidirectional link failures), thus supporting resilient anycast cloud virtual networks. We compare two resilience approaches: PIP-resilience maps each virtual link to two alternate physical routes, VNO-resilience provides alternate paths in the virtual topology (while enforcing physical link disjointness).

/pdf/anycast-end-to-end-resilience-for-cloud-services-over-4c0ervm1bf.pdf

Anycast end-to-end resilience for cloud services over virtual optical networks

Optical networks constitute a fundamental building block that has enabled the success of cloud computing. Virtualization, a cornerstone of cloud computing, today is applied in the networking field: physical network infrastructure is logically partitioned into separate virtual networks, thus providing isolation between distinct virtual network operators (VNOs). Hence, the problem of virtual network mapping has arisen: how to decide which physical resources to allocate for a particular virtual network? In a cloud context, not just network connectivity is required, but also data center (DC) resources located at multiple locations, for computation and/or storage. Given the underlying anycast routing principle, the network operator has some freedom to which specific DC to allocate these resources. In this paper, we solve a resilient virtual network mapping problem that optimally decides on the mapping of both network and multi-location data center resources resiliently using anycast routing, considering time-varying traffic conditions. In terms of resilience, we consider the so-called VNO-resilience scheme, where resilience is provided in the virtual network layer. To minimize physical resource capacity requirements, we allow reuse of both network and DC resources. The failures we protect against include both network and DC resource failures: we hence allocate backup DC resources, and also account for synchronization between primary and backup DC. As optimization criteria, we not only consider resource usage minimization, but also aim to limit virtual network reconfigurations from one time period to the next. We propose a scalable column generation approach to solve the dynamic resilient virtual network mapping problem, and demonstrate it in a case study on a nationwide US backbone network.

http://users.atlantis.ugent.be/cdvelder/papers/2014/bui2014icton.pdf

Time-varying resilient virtual network mapping for multi-location cloud data centers

Multi-layer optical networks have recently evolved towards IP-over-WDM networks. Therein, in order to avoid protection/restoration redundancies against either single or multiple failures, synergies need to be developed between IP and optical layers in order to reduce the costs and the energy consumption of the future IP-over-WDM networks. We propose two new optimization models. The first one is an enhanced cutset model, relying on a column generation reformulation. The second one is a path model, based on a multi-flow formulation. Both models can solve exactly most benchmark instances, which were only solved heuristically so far.

Path vs. Cutset approaches for the design of logical survivable topologies

IP restoration vs. optical protection: Which one has the least bandwidth requirements?

Optical networks are crucial to support increasingly demanding cloud services. Delivering the requested quality of services (in particular latency) is key to successfully provisioning end-to-end services in clouds. Therefore, as for traditional optical network services, it is of utter importance to guarantee that clouds are resilient to any failure of either network infrastructure (links and/or nodes) or data centers. A crucial concept in establishing cloud services is that of network virtualization: the physical infrastructure is logically partitioned in separate virtual networks. To guarantee end-to-end resilience for cloud services in such a set-up, we need to simultaneously route the services and map the virtual network, in such a way that an alternate routing in case of physical resource failures is always available. Note that combined control of the network and data center resources is exploited, and the anycast routing concept applies: we can choose the data center to provide server resources requested by the customer to optimize resource usage and/or resiliency. This paper investigates the design of scalable optimization models to perform the virtual network mapping resiliently. We compare various resilience options, and analyze their compromise between bandwidth requirements and resiliency quality.

/pdf/resilience-options-for-provisioning-anycast-cloud-services-2tjxfmahhp.pdf

Minh N. Bui

Papers

Anycast end-to-end resilience for cloud services over virtual optical networks

Time-varying resilient virtual network mapping for multi-location cloud data centers

Path vs. Cutset approaches for the design of logical survivable topologies

IP restoration vs. optical protection: Which one has the least bandwidth requirements?

Resilience options for provisioning anycast cloud services with virtual optical networks