Ana-Maria Oprescu

Bag-of-Tasks Scheduling under Budget Constraints

Commercial cloud offerings, such as Amazon’s EC2, let users allocate compute resources on demand, charging based on reserved time intervals. While this gives great¿exibility to elastic applications, users lack guidance for choosing between multiple offerings, in order to complete their computations within given budget constraints. In this work, we present BaTS, our budget-constrained scheduler. BaTS can schedule large bags of tasks onto multiple clouds with different CPU performance and cost, minimizing completion time while respecting an upper bound for the budget to be spent. BaTS requires no a-priori information about task completion times, and learns to estimate them at runtime. We evaluate BaTS by emulating different cloud environments on the DAS-3 multi-cluster system. Our results show that BaTS is able to schedule within a user-definedbudget (if such a schedule is possible at all.) At the expense of extra compute time, significant cost savings can be achieved when comparing to a cost-oblivious round-robin scheduler.

Budget Estimation and Control for Bag-of-Tasks Scheduling in Clouds

Commercial cloud offerings, such as Amazons EC2, let users allocate compute resources on demand, charging based on reserved time intervals. While this gives great flexibility to elastic applications, users lack guidance for choosing between multiple offerings, in order to complete their computations within given budget constraints. In this work, we present BaTS, our budget-constrained scheduler. Using a small task sample, BaTS can estimate costs and makespan for a given bag on different cloud offerings. It provides the user with a choice of options before execution and then schedules the bag according to the users preferences. BaTS requires no a-priori information about task completion times. We evaluate BaTS by emulating different cloud environments on the DAS-3 multi-cluster system. Our results show that BaTS correctly estimates budget and makespan for the scenarios investigated; the user-selected schedule is then executed within the given budget limitations.

Dynamic Optimization of SLA-Based Services Scaling Rules

Current advanced cloud infrastructure management solutions allow scheduling actions for dynamically changing the number of running virtual machines (VMs). This approach, however, does not guarantee that the scheduled number of VMs will properly handle the actual user generated workload, especially if the user utilization patterns will change. We propose using a dynamically generated scaling model for the VMs containing the services of the distributed applications, which is able to react to the variations in the number of application users. We answer the following question: How to dynamically decide how many services of each type are needed in order to handle a larger workload within the same time constraints? We describe a mechanism for dynamically composing the SLAs for controlling the scaling of distributed services by combining data analysis mechanisms with application benchmarking using multiple VM configurations. Based on processing of multiple application benchmarks generated data sets we discover a set of service monitoring metrics able to predict critical Service Level Agreement (SLA) parameters. By combining this set of predictor metrics with a heuristic for selecting the appropriate scaling-out paths for the services of distributed applications, we show how SLA scaling rules can be inferred and then used for controlling the runtime scale-in and scale-out of distributed services. We validate our architecture and models by performing scaling experiments with a distributed application representative for the enterprise class of information systems. We show how dynamically generated SLAs can be successfully used for controlling the management of distributed services scaling.

New Instructional Models for Building Effective Curricula on Cloud Computing Technologies and Engineering

This paper presents ongoing work to develop advanced education and training course on the Cloud Computing technologies foundation and engineering by a cooperating group of universities and the professional education partners. The central part of proposed approach is the Common Body of Knowledge in Cloud Computing (CBK-CC) that defines the professional level of knowledge in the selected domain and allows consistent curricula structuring and profiling. The paper presents the structure of the course and explains the principles used for developing course materials, such as Blooms Taxonomy applied for technical education, and andragogy instructional model for professional education and training. The paper explains the importance of using the strong technical foundation to build the course materials that can address interests of different categories of stakeholders and roles/responsibilities in the Cloud Computing services provisioning and operation. The paper provides a short description of summary of the used Cloud Computing related architecture concepts and models that allow consistent mapping between CBK-CC, stakeholder roles/responsibilities and required skills, explaining also importance of the requirements engineering stage that provides a context for cloud based services design. The paper refers to the ongoing development of the educational course on Cloud Computing at the University of Amsterdam, University of Stavanger and provides suggestions for building advanced online training course for IT professionals.

MeTRO: Low latency network paths with routers-on-demand

The current Internet is a loose federation of independent providers (ISPs) That manually manage inter-domain (ASes) route policies To primarily serve Their own interests. The end-user experience may be hindered by Two aspects: The ASes only optimize locally, possibly delivering sub-optimal end-To-end connections; The manual management of routing policies for a large amount of prefixes is error-prone. Infrastructure as a Service (IaaS) clouds let users allocate compute resources on demand, at different geographical locations, while Internet connectivity is guaranteed. Therefore, cloud providers represent untapped resources for a better end-user (application) Internet connectivity experience. In This work we present MeTRO, a framework To construct better Than best-effort routed Internet paths. Our method exploits The fact That cloud computer resources may host virtual routers and That one such router can be part of a path between Two end systems. We perform an extensive evaluation of our method, by deploying it over 75 NLNOG Ring hosts. We show That our method, practically acting as an overlay network, decreases The latency in 58% of The cases studied, albeit increasing The number of hops. Our framework is specifically useful for monitoring and debugging failures, as well as configuration errors related To Internet reachability.

A queueing theory approach to Pareto optimal bags-of-tasks scheduling on clouds

Cloud hosting services offer computing resources which can scale along with the needs of users. When access to data is limited by the network capacity this scalability also becomes limited. To investigate the impact of this limitation we focus on bags–of–tasks where task data is stored outside the cloud and has to be transferred across the network before task execution can commence. The existing bags–of–tasks estimation tools are not able to provide accurate estimates in such a case. We introduce a queuing–network inspired model which successfully models the limited network resources. Based on the Mean–Value Analysis of this model we derive an efficient procedure that results in an estimate of the makespan and the executions costs for a given configuration of cloud virtual machines. We compare the calculated Pareto set with measurements performed in a number of experiments for real–world bags–of–tasks and validate the proposed model and the accuracy of the estimated configurations.

Persistent fault-tolerance for divide-and-conquer applications on the grid

Grid applications need to be fault tolerant, malleable, and migratable. In previous work, we have presented orphan saving, an efficient mechanism addressing these issues for divide-and-conquer applications. In this paper, we present a mechanism for writing partial results to checkpoint files, adding the capability to also tolerate the total loss of all processors, and to allow suspending and later resuming an application. Both mechanisms have only negligible overheads in the absence of faults, even with extremely short checkpointing intervals like one minute. In the case of faults, the new checkpointing mechanism outperforms orphan saving by 10% to 15 %. Also, suspending/resuming an application has only little overhead, making our approach very attractive for writing grid applications.

E-BaTS: Energy-Aware Scheduling for Bag-of-Task Applications in HPC Clusters

High-Performance Computing (HPC) systems consume large amounts of energy. As the energy consumption predictions for HPC show increasing numbers, it is important to make users aware of the energy spent for the execution of their applications. Drawing from our experience with exposing cost and performance in public clouds, in this paper we present a generic mechanism to compute fast and accurate estimates for the tradeoffs between the performance (expressed as makespan) and the energy consumption of applications running on HPC clusters. We validate our approach by implementing it in a prototype, called E-BaTS and validating it with a wide variety of HPC bags-of-tasks. Our experiments show that E-BaTS produces conservative estimates with errors below 5%, while requiring at most 12% of the energy and time of an exhaustive search for providing configurations close to the optimal ones in terms of trade-offs between energy consumption and makespan.

ICOMF: Towards a Multi-cloud Ecosystem for Dynamic Resource Composition and Scaling

Modern cloud-based applications and infrastructures may include resources and services (components) from multiple cloud providers, are heterogeneous by nature and require adjustment, composition and integration. The specific application requirements can be met with difficulty by the current static predefined cloud integration architectures and models. In this paper, we propose the Intercloud Operations and Management Framework (ICOMF) as part of the more general Intercloud Architecture Framework (ICAF) that provides a basis for building and operating a dynamically manageable multi-provider cloud ecosystem. The proposed ICOMF enables dynamic resource composition and decomposition, with a main focus on translating business models and objectives to cloud services ensembles. Our model is user-centric and focuses on the specific application execution requirements, by leveraging incubating virtualization techniques. From a cloud provider perspective, the ecosystem provides more insight into how to best customize the offerings of virtualized resources.

Strategies for Generating and Evaluating Large-Scale Powerlaw-Distributed P2P Overlays

A very wide variety of physical, demographic, biological and man-made phenomena have been observed to exhibit powerlaw behavior, including the population of cities and villages, sizes of lakes, etc. The Internet is no exception to this. The connectivity of routers, the popularity of web sites, and the degrees of World Wide Web pages are only a few examples of measurements governed by powerlaw. The study of powerlaw networks has strong implications on the design and function of the Internet.