Economics of Grids, Clouds, Systems, and Services

The paper focuses the attention to different business models and intended strategic aims of the firms providing Software-as-a-Service (SaaS). SaaS vendors have been said to challenge the business practices of the existing vendors providing proprietary or customer-specific solutions. The current studies on the topic have shown that SaaS is different from preceding software business models, but consider and emphasize SaaS business model as an invariable configuration. This case study compares two SaaS firms with different backgrounds and reveals characteristics of two very different SaaS business models. The findings indicate that along with SaaS vendors providing only standard software applications and focusing on cost efficiency, there are vendors who provide more specialized software applications and complement the SaaS offering with services required by larger customers.

EnergeTIC is a recent industrial research project carried out in Grenoble on optimising energy consumption in data centres. We study the problem formulation proposed by EnergeTIC. The problem focuses on the allocation of virtual machines to servers with time-variable resource demands in data centres in order to minimise energy costs while ensuring service quality. We present a scalable constraint programming-based large neighbourhood search (CP-LNS) method to solving this challenging problem. We present empirical results that demonstrate that the industrial benchmarks can be solved to near optimality using our approach. Our CP-LNS method provides a fast and practical approach for finding high quality solutions for lowering electricity costs in data centres.

Many modern cloud services are provided using Internet Data Centers (IDCs), e.g. the Google search engine. A network of IDCs is implemented using a set of data centers that are geographically distributed over many locations. The energy requirements of these systems are considerable, and there is growing interest in minimizing the total cost of energy required to operate them either by making the hardware more energy efficient or by ensuring that opportunities to access low-cost energy are exploited. In this paper we present a methodology for studying the energy cost implications of minimizing IDC energy costs under different operational and energy cost prediction regimes. We systematically study the impact of the level of price variability, time lag between locations due to the geographical distribution, reconfiguration delay, and accuracy of price predictions, on the overall electricity cost associated with managing an IDC.

Cloud computing is considered to be energy and ecological efficient, and is promoted as the environmental friendly computing solution. On the other hand, the massive development of the Cloud marketplace lead in an increase of the Data Centers globally and eventually in the increase of the CO₂ related footprint. The calculation of the impact of Virtual Machines (VMs) on the environment is a challenging task, not only due to the technical difficulties but also due to the lack of information from the energy providers. In this paper we present a methodology for the estimation of the ecological efficiency of Virtual Machines in Cloud infrastructures. We focus on the information management in relation with the energy production in a region as well as the ecological efficiency of a VM in a Data Center. To this end, we have designed and implemented a framework through which the ecological efficiency can be monitored. The presented framework is being evaluated through a private Cloud scenario deployed into infrastructure located in Germany.

Cloud has emerged as an alternative to clusters and grids. Its adoption as an execution environment capable of supporting the high requirements of scientific computations is still an open question. In a previous work, the authors conducted successfully a practical experience of taking advantage of clusters and grids to solve a semantic annotation problem in 178 days. In this work, the authors analyse the cost of solving that problem and compare it with the cost of solving it in a pure Cloud scenario. For this last, a detailed cost estimation is conducted according to the data extracted from the actual execution of a reduced dataset on Amazon EC2. As a result, the suitability of using Cloud-based solutions to solve large and complex scientific problems is discussed.

Large scale clusters are now being used in shared, multi-tenant scenarios by heterogeneous applications with completely different requirements. In this scenario, it’s useful to explore the intersection of two complementary goals. On one side, energy efficiency is an important factor to consider in this world with increasing operating costs related to energy consumption. On the other side, heterogeneous applications emphasize the problem of distributing the execution capacity among competitive users in a shared setting. In this paper, we address the combination of these two goals by introducing an incentive mechanism to make users report their actual resource requirements, allowing them to dynamically scale-up or down as necessary. In turn, this information is used by the infrastructure operator to shut down resources without reducing the QoS provided to users and effectively reducing energy costs. We show how our mechanism is able to meet the performance requirements of applications without over-provisioning physical resources, which in turn translates into energy savings.

With the increasing availability of streaming applications from mobile devices to dedicated sensors, understanding how such streaming content can be processed within some time threshold remains an important requirement. We investigate how a computational infrastructure responds to such streaming content based on the revenue per stream – taking account of the price paid to process each stream, the penalty per stream if the pre-agreed throughput rate is not met, and the cost of resource provisioning within the infrastructure. We use a token-bucket based rate adaptation strategy to limit the data injection rate of each data stream, along with the use of a shared token-bucket to enable better allocation of computational resource to each stream. We demonstrate how the shared token-bucket based approach can enhance the performance of a particular class of applications, whilst still maintaining a minimal quality of service for all streams entering the system.

Efficient load distribution plays an important role in grid and cloud applications. In a typical problem, a divisible load should be split into parts and allocated to several processors, with one processor responsible for the data transfer. Since processors have different speed and cost characteristics, selecting the processor order for the transmission and defining the chunk sizes affect the computation time and cost. We perform a systematic study of the model analysing the properties of Pareto optimal solutions. We demonstrate that the earlier research has a number of limitations. In particular, it is generally assumed that the load should be distributed so that all processors have equal completion times, while in fact this property is satisfied only for some deadlines; for many optimal schedules this property does not hold. Moreover, fixing the processor sequence in the non-decreasing order of the cost-characteristic may be appropriate only for Pareto-optimal solutions with relatively large deadlines; optimal schedules for tight deadlines may have a different order of processors. We conclude with an efficient algorithm for finding the time-cost trade-off.

The allocation of resources between providers to consumers is a well-known problem and has received significant attention, typically using notions of monetary exchanges. In this paper, we study resource matching in settings without monetary transactions by using a two-sided matching approach, e.g., in social and collaborative environments where users define preferences for with whom they may be matched. Whereas two-sided matching for strict and complete preference rankings (i.e., without indifferences) has been extensively studied, it is known that the matching problem is NP-hard for more realistic preference structures. We study, via simulation, the applicability of a heuristic procedure in settings with indiffernces in preferences, and compare its performance to existing algorithms. We study performance metrics like fairness and welfare in addition to the classic stability objective. Our results show interesting trade-offs between performance metrics and promising performance of the heuristic.

Efficiently scheduling tasks in hybrid Distributed Computing Infrastructures (DCI) is a challenging pursue because the scheduler must deal with a set of parameters that simultaneously characterize the tasks and the hosts originating from different types of infrastructure.

In this paper we propose a scheduling method for hybrid DCIs, based on advanced multi-criteria decision methods. The scheduling decisions are made using pairwise comparisons of the tasks for a set of criteria like expected completion time and price charged for computation. The results are obtained with an XtremWeb-like pull-based scheduler simulator using real failure traces from [1] for a combination of three types of infrastructure. We also show how such a scheduler should be configured to enhance user satisfaction regardless their profiles, while maintaining good values for makespan and cost.

We validate our approach with a statistical analysis on empirical data and show that our proposed scheduling method improves performance by 12-17% compared to other scheduling methods. Experimenting on large time-series and using realistic scheduling scenarios lead us to conclude about time consistency results of the method.

Cloud computing has the potential to improve resource efficiency by consolidating many virtual computers onto each physical host. This economization is based on the assumption that a significant percentage of virtual machines are indeed not fully utilized. Yet, despite the much acclaimed pay-only-for-what-you-use paradigm, public IaaS cloud customers are usually still billed by the hour for virtual systems of uncertain performance rather than on the basis of actual resource usage. Because ensuring and proving availability of defined performance for collocated multi-tenant VMs poses a complex technical problem, providers are still reluctant to provide performance guarantees. In lack thereof, prevailing cloud products range in the low price segment, where providers resort to overbooking and double selling capacity in order to maintain profitability, thereby further harming trust and cloud adoption. In this paper we argue that the predominant flat rate billing in conjunction with the practice of overbooking and its associated mismatch between actual costs and billed posts results in a substantial misalignment between the interests of providers and customers that stands in the way of trustworthy and sustainable cloud computing. On these grounds, we propose a hybrid IaaS pricing model that aims to avoid these problems in a non-technical fashion by shifting to consumption based billing on top of credible minimum performance. Requiring only measures that can be obtained with a low degree of technical complexity as well as a moderate amount of trust, the approach aspires to be more sustainable, practicable and billable than common practice even without the use of complex should-I verifiability.

With the increase in computing infrastructure commercialization through the pay-as-you-go model, competition among providers puts the user as a decision agent on which is the best provider to comply with his/her demands and requirements. Currently, users rely on instances offered as on-demand, reserved, and spot to decide which is the best resource allocation model over the time. In this work, we present substantial contributions to compose a PaaS architecture that leverages different charging models, where we propose the use of a new charging model called time-slotted reservation. Moreover, we developed an integer linear program (ILP) to perform the scheduling of incoming requests according to different QoS levels, proposing a mapping of those levels into the charging models offered by IaaS providers. Simulations show the applicability of the ILP in the proposed model, being able to maximize the number of requisitions executed following the user’s QoS requirements.

Community networks are built with off-the-shelf communication equipment aiming to satisfy a community’s demand for Internet access and services. These networks are a real world example of a collective that shares ICT resources. But while these community networks successfully achieve the IP connectivity over the shared network infrastructure, the deployment of applications inside of community networks is surprisingly low. Given that community networks are driven by volunteers, we believe that bringing in incentive-based mechanisms for service and application deployments in community networks will help in unlocking its true potential. We investigate in this paper such mechanisms to steer user contributions, in order to provide cloud services from within community networks. From the analysis of the community network’s topology, we derive two scenarios of community clouds, the local cloud and the federated cloud. We develop an architecture tailored to community networks which integrates the incentive mechanism we propose. In simulations of large scale community cloud scenarios we study the behaviour of the incentive mechanism in different configurations, where slices of homogeneous virtual machine instances are shared. Our simulation results allow us to understand better how to configure such an incentive mechanism in a future prototype of a real community cloud system, which ultimately should lead to realisation of clouds in community networks.

In recent years, the rise and rapid adoption of cloud computing has acted as a catalyst for research in related fields: virtualization, distributed and service-oriented computing to name but a few. Whilst cloud computing technology is rapidly maturing, many of the associated long-standing socio-technical challenges including the dependability of cloud-based service composition, services manageability and interoperability remain unsolved. These can be argued to slow down the migration of serious business critical applications to the cloud model. This paper reports on progress towards the development of a method to generate cloud-based service compositions from requirements metadata. The paper presents a formal approach that uses Situation Calculus to translate service requirements into an Intention Workflow Model (IWM). This IWM is then used to generate autonomic cloud service composition. The Petshop benchmark is used to illustrate and evaluate the proposed method.

This paper aims to highlight the importance of End-to-End (E2E) service quality for cloud scenarios, with focus on telecom carrier-grade services. In multi-tenant distributed and virtualized cloud infrastructures, enhanced resource sharing raises issues in terms of performance stability and reliability. Moreover, the heterogeneity of business entities responsible for the cloud service delivery, threatens the possibility of offering precise E2E service levels.

Setting up proper Service-Level Agreements (SLAs) among the involved players, may become overly challenging. However, problems may be mitigated by a thoughtful intervention of standardization.

The paper reviews some of the most important efforts in research and industry to tackle E2E service quality and concludes with some recommendations for additional research and/or standardization effort required to be able to deploy mission critical or interactive real-time services with high demands on service quality, reliability and predictability on cloud platforms.

Service markets allow users to discover, purchase, and utilize services offered on a specific platform. As service platforms grow in number of users and variety of offerings, it raises the question of whether this phenomenon continues to benefit users. Based on a literature review, the paper identifies usability, service variety, and the number of personal connections accessible over the service platform as major determinants that contribute to the value to users. Based on survey data on the behavior of mobile service users, the relationship between user value and the determinants is analyzed and estimated. The results show positive correlations between all three determinants and the value. Using regressions, we estimate how much these determinates contribute to the user value. Mobile service users are satisfied with the usability of services of their chosen platforms, although the impact on the user value is the lowest. Users benefit the most from an increase in the number of their personal connections and the number of services they use.

The Learned Hand’s rule, comparing security investments against the expected loss from data breaches, can be used as a simple tool to determine the negligence of the company holding the data. On the other hand, companies may determine their investments in security by maximizing their own net profit. We consider the well known Gordon-Loeb models as well as the more recent Huang-Behara models for the relationship between investments and the probability of money loss due to malicious attacks to determine the outcome of the application of three forms of Hand’s rule: status quo (loss under no investments), ex-post (loss after investment), transitional (loss reduction due to investment). The company is always held negligent if it does not invest in both the status quo and the transitional form. In the ex-post form, it is instead held negligent just if the potential loss is below a threshold, for which we provide the exact expression.