Di Valerio

A game-theoretic approach to computation offloading in mobile cloud computing

We consider a three-tier architecture for mobile and pervasive computing scenarios, consisting of a local tier of mobile nodes, a middle tier (cloudlets) of nearby computing nodes, typically located at the mobile nodes access points but characterized by a limited amount of resources, and a remote tier of distant cloud servers, which have practically infinite resources. This architecture has been proposed to get the benefits of computation offloading from mobile nodes to external servers while limiting the use of distant servers whose higher latency could negatively impact the user experience. For this architecture, we consider a usage scenario where no central authority exists and multiple non-cooperative mobile users share the limited computing resources of a close-by cloudlet and can selfishly decide to send their computations to any of the three tiers. We define a model to capture the users interaction and to investigate the effects of computation offloading on the users’ perceived performance. We formulate the problem as a generalized Nash equilibrium problem and show existence of an equilibrium. We present a distributed algorithm for the computation of an equilibrium which is tailored to the problem structure and is based on an in-depth analysis of the underlying equilibrium problem. Through numerical examples, we illustrate its behavior and the characteristics of the achieved equilibria.

Optimal Pricing and Service Provisioning Strategies in Cloud Systems: A Stackelberg Game Approach

In this paper we consider several Software as a Service (SaaS) providers, that offer a set of applications using the Cloud facilities provided by an Infrastructure as a Service (IaaS) provider. We assume that the IaaS provider offers a pay only what you use scheme similar to the Amazon EC2 service, comprising flat, on demand, and spot virtual machine instances. We propose a two stage provisioning scheme. In the first stage, the SaaS providers determine the number of required flat and on demand instances by means of standard optimization techniques. In the second stage the SaaS providers compete, by bidding for the spot instances which are instantiated using the unused IaaS capacity. We assume that the SaaS providers want to maximize a suitable utility function which accounts for both the QoS delivered to their users and the associated cost. The IaaS provider, on the other hand, wants to maximize his revenue by determining the spot prices given the SaaS bids. We model the second stage as a Stackelberg game, and we compute its equilibrium price and allocation strategy by solving a Mathematical Program with Equilibrium Constraints (MPEC) problem. Through numerical evaluation we study the equilibrium solutions as function of the system parameters.

A Scalable and Highly Available Brokering Service for SLA-Based Composite Services

The introduction of self-adaptation and self-management techniques in a service-oriented system can allow to meet in a changing environment the levels of service formally defined with the system users in a Service Level Agreement (SLA). However, a self-adaptive SOA system has to be carefully designed in order not to compromise the system scalability and availability. In this paper we present the design and performance evaluation of a brokering service that supports at runtime the self-adaptation of composite services offered to several concurrent users with different service levels. To evaluate the performance of the brokering service, we have carried out an extensive set of experiments on different implementations of the system architecture using workload generators that are based on open and closed system models. The experimental results demonstrate the effectiveness of the brokering service design in achieving scalability and high availability.

A performance comparison of QoS-driven service selection approaches

Service selection has been widely investigated as an effective adaptation mechanism that allows a service broker, offering a composite service, to bind each task of the abstract composition to a corresponding implementation, selecting it from a set of candidates. The selection aims typically to fulfill the Quality of Service (QoS) requirements of the composite service, considering several QoS parameters in the decision. We compare the performance of two representative examples of the perrequest and per-flow approaches that address the service selection issue at a different granularity level. We present experimental results obtained with a prototype implementation of a service broker. Our results show the ability of the per-flow approach in sustaining an increasing traffic of requests, while the per-request approach appears more suitable to offer a finer customizable service selection in a lightly loaded system.

A new approach to QoS driven service selection in service oriented architectures

Service selection has been widely investigated by the SOA research community as an effective adaptation mechanism that allows a service broker, offering a composite service, to bind at runtime each task of the composite service to a corresponding concrete implementation, selecting it from a set of candidates which differ from one another in terms of QoS parameters. In this paper we present a load-aware per-request approach to service selection which aims to combine the relative benefits of the well known per-request and per-flow approaches. We present experimental results obtained with a prototype implementation of a service broker. Our results show that the proposed approach is superior to the traditional per-request one and combines the ability of sustaining large volume of service requests, as the per-flow approach, while at the same time offering a finer customizable service selection, as the per-request approach.

A self-adaptive protocol stack for Underwater Wireless Sensor Networks

In this paper we propose the adoption of a self-adaptable cross-layer and modular Software Defined Communication Stack (SDCS) for Underwater Wireless Sensor Networks. The SDCS is a modular stack solution which is capable to run different protocols at each layer of the network stack; a new component, named policy engine, autonomously and adaptively, as the operational conditions vary, selects the protocol of each layer so as to optimize application scenario metrics of interest, e.g., packet delivery ratio, end-to-end packet latency and energy consumption. As a proof of concept, the paper presents the design and performance evaluation of a policy engine to dynamically and autonomously change the MAC protocol adopted in Underwater Wireless Sensor Networks. The best MAC protocol is chosen according to network conditions and application requirements, without any a priori knowledge. We consider three different MAC protocols running in the SDCS: CSMA, T-Lohi and DACAP that represent the class of simple, intermediate and fully negotiated MAC protocols, respectively. The performance of the three protocols are first compared via simulations considering different network conditions, such as traffic load and packet size. Then, we evaluate the ability of our policy engine to dynamically estimate the network changes and then to select accordingly the best MAC protocol without any a priori knowledge. Results show the effectiveness of our solution in that it is always able to quickly find and choose the MAC protocol that optimizes a given metric in a particular scenario by introducing a really limited overhead in the network.

Optimal Virtual Machines allocation in mobile femto-cloud computing: An MDP approach

Offloading to external surrogate machines (part of) the workload generated by applications running on mobile nodes has been suggested as a way to improve the mobile user experience. In this paper, we consider a set of mobile users that can offload their computation on Virtual Machines (VMs) instantiated in a cloud infrastructure implemented over a set of femtocells which have been augmented with computational resources. In this setting, a critical task is to determine where users VMs should be allocated across the different femtocells as to optimize the user performance. In this paper we formulate the VMs allocation problem as a Markov Decision Process (MDP) based optimization, to directly take into account the system dynamics and optimize the long term performance. The optimal policy is obtained solving the MDP using a Linear Programming reformulation. We illustrate the behavior of the proposed policy in simple scenarios.

Bidding Strategies in QoS-Aware Cloud Systems Based on N-Armed Bandit Problems

In this paper we consider a set of Software as a Service (SaaS) providers, that offer a set of Web services using the Cloud facilities provided by an Infrastructure as a Service (IaaS) provider. We assume that the IaaS provider offers a pay only what you use scheme similar to the Amazon EC2 service, comprising flat, on demand, and spot virtual machine instances. We propose a two-stage provisioning scheme. In the first stage, the SaaS providers determine the number of required flat and on demand instances by means of standard optimization techniques. In the second stage, the SaaS providers compete by bidding for the spot instances which are instantiated using the unused IaaS capacity. We put our focus on the bidding decision process by the SaaS providers, which takes place during the second stage, and apply N-armed bandit problems, in which the player is faced repeatedly with a choice among N different options, and every time he submits his decision evaluating past feedbacks. Through numerical experiments, we analyze proposed strategies under different scenarios and prove the SaaS providers ability to refine their behavior round by round and to determine the best bid so to maximize their revenue and achieve as many spot resources as possible, also addressing the importance of a trade-off between exploration and exploitation, i.e., among greedy and non-greedy actions.

Finding MARLIN: Exploiting multi-modal communications for reliable and low-latency underwater networking

This paper concerns the smart exploitation of multimodal communication capabilities of underwater nodes to enable reliable and swift underwater networking. To contrast adverse and highly varying channel conditions we define a smart framework enabling nodes to acquire knowledge on the quality of the communication to neighboring nodes over time. Following a model-based reinforcement learning approach, our framework allows senders to select the best forwarding relay for its data jointly with the best communication device to reach that relay. We name the resulting forwarding method MARLIN, for MultimodAl Reinforcement Learning-based RoutINg. Applications can choose whether to seek reliable routes to the destination, or whether faster packet delivery is more desirable. We evaluate the performance of MARLIN in varying networking scenarios where nodes communicate through two acoustic modems with widely different characteristics. MARLIN is compared to state-of-the-art forwarding protocols, including a channel-aware solution, a machine learning-based solution and to a flooding protocol extended to use multiple modems. Our results show that a smartly learned selection of relay and modem is key to obtain a packet delivery ratio that is twice as much that of other protocols, while maintaining low latencies and energy consumption.

WHARP: A Wake-Up Radio and Harvesting-Based Forwarding Strategy for Green Wireless Networks

Green wireless networks are characterized by devices that are pervasively deployed and that harvest energy from the surrounding environment. Devices are also endowed with low-power triggering techniques (e.g., wake-up radios) to obviate costly idle communication times. In this paper, we present a novel data forwarding strategy for green wireless networks that fully exploits the self-powered wake-up radio capabilities of the network nodes. The proposed strategy, named WHARP for Wake-up and HARvesting-based energy-Predictive forwarding, sends data to their destination by making decentralized and proactive decisions based on forecast energy and expected traffic. The performance of WHARP has been compared to that of the Energy Harvesting Wastage-Aware (EHWA) strategy through GreenCastalia-based simulations. Results show that our approach delivers up to 72% more packets, 1.6 times faster, and consuming 58% less energy than EHWA. This is obtained through a learned selection of forwarder relays allowing WHARP nodes to be operational 98% of the time: A 30% improvement over EHWA.