Daniel Lázaro

A simulation-optimization approach to deploy Internet services in large-scale systems with user-provided resources

Cloud computing systems can benefit from the use of personal and non-dedicated computers, which are currently employed in volunteer computing systems. Being non-dedicated, these resources show random behavior regarding the times they are online (available) and offline. Accordingly, their availability levels are lower than those of traditionally employed dedicated resources. Thus, in order to use non-dedicated resources in cloud computing environments, it is necessary first to solve the problem of how to attain high availability levels for the Internet services deployed over them. Most approaches on how to guarantee high service availability levels with non-dedicated resources are based on the introduction of high degrees of redundancy into the system. However, this praxis leads to an inefficient usage of computational resources and, therefore, to higher operational costs. Accordingly, the focus of this paper is the problem of minimizing the cost of a service deployment over non-dedicated resources while providing a high level of service availability. In order to solve this stochastic optimization problem, the paper proposes a hybrid algorithm that combines a metaheuristic component with a discrete-event simulation component. The metaheuristic component is used to search for an efficient configuration of resources. The simulation component is integrated inside the metaheuristic and used to estimate the service availability of each promising configuration. A numerical experiment section, comparing the performance of several algorithms, contributes to validating the proposed approach as well as to illustrate its potential applications.

Decentralized resource discovery mechanisms for distributed computing in peer-to-peer environments

Resource discovery is an important part of distributed computing and resource sharing systems, like grids and utility computing. Because of the increasing importance of decentralized and peer-to-peer environments, characterized by high dynamism and churn, a number of resource discovery mechanisms, mainly based on peer-to-peer techniques, have been presented recently. We present and classify them according to criteria like their topology and the degree of achievement of various common requirements of great importance for the targeted environments, as well as compare their reported performance. These classifications intend to provide an intuitive vision of the strengths and weaknesses of each system.

DyMRA: dynamic market deployment for decentralized resource allocation

The workload supported by Virtual Organizations (VO) is limited by the quantity of available resources. VOs with scarce resources or peer-to-peer based VOs--due to the dynamicity of available resources -- may need extra resources to carry out a given task. Conversely, many Internet-connected computers have surplus bandwidth, storage and computational resources. We face those tradeoffs by enabling VOs to collect and aggregate surplus resources and provide them with availability guarantees to other VOs. This paper presents DyMRA, a decentralized resource allocation system based on markets that allows inter-VO resource allocation. DyMRA is specially designed for dynamic and peer-to-peer environments, where the autonomy of participants to disconnect resources at any time and its decentralized nature requires the capacity to dynamically reallocate resources and services that manage the overall system. DyMRA is built on top of LaCOLLA, a peer-to-peer middleware that allows a group of users to share resources in a collaborative manner. We present the design, architecture and validation of our proposal.

Flexible Resource Discovery for Decentralized P2P and Volunteer Computing Systems

Resource discovery mechanisms are key for distributed computing. Traditional approaches use dedicated servers to store information about the whole system and make all the scheduling decisions. Many mechanisms exist for decentralized environments, based on distributing the information in a certain way over a structured overlay network. This usually creates complex relationships between the distribution of the data to be stored and the performance of the system. In this paper, we present a method for resource discovery with the simplicity and power of centralized approaches, designed for decentralized systems formed by non-dedicated resources. We show that it provides a constant look-up latency for frequent resources in the average case, while causing a low overhead and tolerating the levels of churn that are to be expected from non-dedicated resources.

An Architecture for Decentralized Service Deployment

In this paper we present a proposal of the architecture for a system which allows the deployment of services in a group of computers, connected in a peer-to-peer fashion. This architecture is divided in layers, and each of them contains some components which offer specific functions. By putting them together, we obtain a system with desirable characteristics such as scalability, decentralization, ability to deal with heterogeneity, fault tolerance, load-balancing, and self-* properties.

Towards Decentralized Resource Allocation for Collaborative Peer to Peer Learning

Collaborative e-learning virtual communities use virtual learning environments provided by the university or tools that are available in Internet. Although this model has proven its feasibility, it has important limitations. A way to deal with them is by defining virtual organizations (VO) that gather the resources and interests of their members in a way that they can lend resources to or borrow them from other VO. This paper presents DyMRA, a decentralized resource allocation system based on markets that allows inter-VO resource allocation. DyMRA is specially designed for collaborative peer-to-peer environments, where the autonomy of participants and its decentralized nature requires the capacity to dynamically reallocate resources and services that manage the overall system. DyMRA is built on top of LaCOLLA, a peer-to-peer middleware that allows a group of users to share resources in a collaborative manner. We present the design, architecture and validation of our proposal.

Towards an architecture for service deployment in contributory communities

This paper proposes an architecture and a set of mechanisms for deploying services in a decentralised contributory system, i.e. a system that runs using only the resources contributed to the community by its members. We present the layers that form our architecture and show the existing technologies that can be used to implement some of them. We show how our mechanisms make use of these layers and how they interact to allow the deployment of services in a decentralised way. The resources contributed by the members of the community are used to host the services deployed in it, while achieving desirable characteristics such as scalability, decentralisation, ability to deal with heterogeneity, fault tolerance, load-balancing and self-properties.

DyMRA: A Decentralized Resource Allocation Framework for Collaborative Learning Environments

Collaborative e-learning virtual communities use virtual learning environments provided by the university or tools that are available in Internet. Although this model has proven to work, it has important limitations. A way to deal with them is by defining virtual organizations (VO) that gather the resources and interests of their members in a way that they can lend resources to or borrow them from other VO. This paper presents DyMRA, a decentralized resource allocation system based on markets that allows inter-VO resource allocation. DyMRA is specially designed for collaborative peer-to-peer environments, where the autonomy of participants and its decentralized nature requires the capacity to dynamically reallocate resources and services that manage the overall system. DyMRA is built on top of La- COLLA, a peer-to-peer middleware that allows a group of users to share resources in a collaborative manner. We present the design, architecture and validation of our proposal.

A Multi-lane Double Auction for Economic-Based Service Management in the Cloud

Economic models have shown their suitability to allocate resources efficiently, considering an unbalanced supply and demand. As the use of the Cloud is extending, a numerous set of distributed resource allocation frameworks have been developed to attain efficient resource management while keeping the scalability of the infrastructure. However, those frameworks make use of either simple double auction mechanisms or complex approximations to the NP-complete problem of the combinatorial auction. The problem of those mechanisms is that of its generality, that is, they have not been specially designed for the trading of time-leased computational resources. In this paper we present a novel variant of the double auction that has been specially adapted to trade time-differentiated items as Cloud services can be considered. The paper presents the data structures, algorithms and architecture of the economic mechanism as well as it presents the evaluation of the mechanism through simulation. Simulated results are compared with the main double auction implementations found in the literature. The paper constitutes an approach to improve efficiency of service management and allocation in the Cloud from the point of view of the economic model and not from architectural aspects addressed by most of the contributions found in the literature.

A self-* auction server: design principles, architecture and implementation

The use of economic models to handle resource allocation in the grid is a reality. Auction mechanisms are used in grid and distributed testbeds to elicitate users preferences while improving the performance of the system. Existing implementations rely on single auction mechanisms to allocate resources whilst grids are heterogeneous in nature. Many different applications cohabitate having different workflow requirements such as bag of tasks executions or real time interaction that cannot be dealt efficiently by a single mechanism. In that scenario it would be more realistic to allocate resources using the most suitable mechanism for each situation, enabling also its configuration, deployment and management at runtime. Therefore, the paper presents the design principles, architecture and implementation of a configurable auction server (CAS). The auction server offers support for the deployment, configuration and execution of different auction mechanisms, facilitating the task of market mechanism developers and enabling the execution of distributed marketplaces according to local demand.