A5000119692- Distributed and Parallel Computing Systems
- Parallel Computing and Optimization Techniques
- Cloud Computing and Resource Management
- Scientific Computing and Data Management
- Advanced Data Storage Technologies
- Software System Performance and Reliability
- Distributed systems and fault tolerance
- Peer-to-Peer Network Technologies
- Embedded Systems Design Techniques
- IoT and Edge/Fog Computing
- Evolutionary Algorithms and Applications
- Service-Oriented Architecture and Web Services
- Software Engineering Research
- Research Data Management Practices
- Interconnection Networks and Systems
- Software Testing and Debugging Techniques
- Reservoir Engineering and Simulation Methods
- Simulation Techniques and Applications
- Algorithms and Data Compression
- Business Process Modeling and Analysis
- Advanced Software Engineering Methodologies
- Computational Physics and Python Applications
- Scheduling and Optimization Algorithms
- Model-Driven Software Engineering Techniques
- Caching and Content Delivery
Universität Innsbruck
2016-2025
University of International Business
2021
Software (Spain)
2020
Pillar (Ukraine)
2020
Inform (Germany)
2018
Friedrich-Alexander-Universität Erlangen-Nürnberg
2018
University of Vienna
1997-2014
Management Center Innsbruck
2014
Czech Academy of Sciences, Institute of Computer Science
2006
Institut de Recherche en Informatique et Systèmes Aléatoires
2005
Cloud computing is an emerging commercial infrastructure paradigm that promises to eliminate the need for maintaining expensive facilities by companies and institutes alike. Through use of virtualization resource time sharing, clouds serve with a single set physical resources large user base different needs. Thus, have potential provide their owners benefits economy scale and, at same time, become alternative scientists clusters, grids, parallel production environments. However, current been...
Workflows have emerged as a paradigm for representing and managing complex distributed computations are used to accelerate the pace of scientific progress. A recent National Science Foundation workshop brought together domain, computer, social scientists discuss requirements future applications challenges they present current workflow technologies.
Scheduling is a key concern for the execution of performance-driven Grid applications. In this paper we comparatively examine different existing approaches scheduling scientific workflow applications in environment. We evaluate three algorithms namely genetic, HEFT, and simple "myopic" compare incremental partitioning against full-graph strategy. demonstrate experiments using real-world covering both balanced (symmetric) unbalanced (asymmetric) workflows. Our results that with HEFT algorithm...
Abstract Performance engineering of parallel and distributed applications is a complex task that iterates through various phases, ranging from modeling prediction, to performance measurement, experiment management, data collection, bottleneck analysis. There no evidence so far all these phases should/can be integrated into single monolithic tool. Moreover, the emergence computational Grids as common wide‐area platform for high‐performance computing raises idea provide tools interacting Grid...
We present the ASKALON environment whose goal is to simplify development and execution of workflow applications on Grid. centered around a set high-level services for transparent effective Grid access, including Scheduler optimized mapping workflows onto Grid, an Enactment Engine reliable application execution, Resource Manager covering both computers components, Performance Prediction service based training phase statistical methods. A sophisticated XML-based programming interface that...
Traditional scheduling research usually targets make span as the only optimization goal, while several isolated efforts addressed problem by considering at most two objectives. In this paper we propose a general framework and heuristic algorithm for multi-objective static of scientific workflows in heterogeneous computing environments. The uses constraints specified user each objective approximates optimal solution applying double strategy: maximizing distance to constraint vector dominant...
The ultimate goal of cloud providers by providing resources is increasing their revenues. This leads to a selfish behavior that negatively affects the users commercial multicloud environment. In this paper, we introduce pricing model and truthful mechanism for scheduling single tasks considering two objectives: monetary cost completion time. With respect social mechanism, i.e., minimizing time cost, extend dynamic scientific workflows. We theoretically analyze truthfulness efficiency present...
In Infrastructure as a Service (IaaS) Clouds, users are charged to utilize cloud services according pay-per-use model. If intend run their workflow applications on resources within specific budget, they have adjust demands for with respect this budget. Although several scheduling approaches introduced solutions optimize the makespan of workflows set heterogeneous IaaS certain hourly-based cost model some well-known providers (e.g., Amazon EC2 Cloud) can easily lead higher and schedulers may...
Task-based programming models for shared memory—such as Cilk Plus and OpenMP 3—are well established documented. However, with the increase in parallel, many-core, heterogeneous systems, a number of research-driven projects have developed more diversified task-based support, employing various runtime features. Unfortunately, despite fact that dozens different systems exist today are actively used parallel high-performance computing (HPC), no comprehensive overview or classification...
Many techniques such as scheduling and resource provisioning rely on performance prediction of workflow tasks for varying input data. However, estimates are difficult to generate in the cloud. This paper introduces a novel two-stage machine learning approach predicting task execution times data In order achieve high accuracy predictions, our relies parameters reflecting runtime information two stages predictions. Empirical results four real world applications several commercial cloud...
Currently grid application developers often configure available components into a workflow of tasks that they can submit for executing on the grid. In this paper, we present an abstract language (AGWL) describing applications at high level abstraction. AGWL has been designed such user concentrate specifying without dealing with either complexity or any specific implementation technology (e.g. Web service). is XML-based which allows programmer to define graph activities refer mostly...
Advance reservation of Grid resources can play a key role in enabling middleware to deliver on-demand resource provision with significantly improved Quality-of-Service (QoS). However, the Grid, advance has been largely ignored due dynamic behavior, under-utilization concerns, multi-constrained applications, and lack support for agreement enforcement. These issues force make allocations at runtime reduced QoS. To remedy these, we introduce new, 3-layered negotiation protocol resources. We...
Unleashing the full potential of heterogeneous systems, consisting multi-core CPUs and GPUs, is a challenging task due to difference in processing capabilities, memory availability, communication latencies different computational resources.
The drift towards new challenges in grid computing, including the utility paradigm and service level agreements based on quality-of-service guarantees, implies need for new, robust, multi-criteria scheduling algorithms that can be applied by user an intuitive way. Multiple criteria addressed related research include execution time, cost of running a task machine, reliability, different data quality metrics. existing bi-criteria approaches are usually dedicated certain criterion pairs only...
In this paper we introduce a multi-objective autotuning framework comprising compiler and runtime components. Focusing on individual code regions, our uses novel search technique to compute set of optimal solutions, which are encoded into multi-versioned executable. This enables the system choose specifically tuned versions when dynamically adjusting changing circumstances. We demonstrate method by tuning loop tiling in cache-sensitive parallel programs, optimizing for both efficiency. Our...
Task clustering has proven to be an effective method reduce execution overhead and improve the computational granularity of scientific workflow tasks executing on distributed resources. However, a job composed multiple may have higher risk suffering from failures than single task job. In this paper, we conduct theoretical analysis impact transient runtime performance executions. We propose general failure modeling framework that uses maximum likelihood estimation-based parameter estimation...
Serverless workflow applications or function choreographies (FCs), which connect serverless functions by data- and control-flow, have gained considerable momentum recently to create more sophisticated as part of Function-as-a-Service (FaaS) platforms. Initial experimental analysis the current support for FCs uncovered important weaknesses, including provider lock-in, limited data-flow control-flow constructs. To overcome some these we introduce Abstract Function Choreography Language (AFCL)...
The existing Grid workflow scheduling projects do not handle recursive loops which are characteristic to many scientific problems. We propose a hybrid approach for Directed Graph (DG)-based workflows in environment with dynamically changing computational and network resources. Our dynamic algorithm is based on the iterative invocation of classical static Acyclic Graphs (DAGs) heuristics generated using well-defined cycle elimination task migration techniques. problem as an application...