9th International Workshop on Middleware for Grids, Clouds and e-Science - MGC 2011


In conjunction with ACM/IFIP/USENIX 12th International Middleware Conference 2011

Lisbon, Portugal. December 12, 2011

Banner 1 Banner 2

Technical Program



Time Title/Autors
14:15 Opening Session
14:30 Prediction-based Auto-scaling of Scientific Workflows
Reginald Cushing (Informatics Institute, University of Amsterdam, Netherlands)
Spiros Koulouzis (Informatics Institute, University of Amsterdam, Netherlands)
Adam S. Z. Belloum (Informatics Institute, University of Amsterdam, Netherlands)
15:00 Power-Aware Virtual Machine Scheduling on Clouds Using Active Cooling Control and DVFS
Daniel Guimaraes do Lago (Institute of Computing, State University of Campinas - UNICAMP, Brazil)
Edmundo R. M. Madeira (Institute of Computing, State University of Campinas - UNICAMP, Brazil)
Luiz Fernando Bittencourt (Institute of Computing, State University of Campinas - UNICAMP, Brazil)
15:30 Understanding Scheduling Implications for Scientific Applications in Clouds
Giacomo Mc Evoy (Distributed Scientific Computing (ComCiDis), National Laboratory for Scientific Computing - LNCC, Brazil)
Bruno Schulze (Distributed Scientific Computing (ComCiDis), National Laboratory for Scientific Computing - LNCC, Brazil)
16:00 Coffee Break (Posters Presentation)
16:30 A Hybrid Fault Tolerance Scheme for EasyGrid MPI Applications
Jacques A. da Silva (Institute of Computing, Federal Fluminense University - UFF, Brazil)
Vinod E. F. Rebello (Institute of Computing, Federal Fluminense University - UFF, Brazil)
17:00 RvfcBOX: Multi-User Consistent File Sharing
Jean-Pierre Ramos (INESC ID / Technical University of Lisbon, Portugal)
Luis Veiga (INESC ID / Technical University of Lisbon, Portugal)
Paulo Ferreira (INESC ID / Technical University of Lisbon, Portugal)
Abstracts
Prediction-based Auto-scaling of Scientific Workflows
Reginald Cushing (Informatics Institute, University of Amsterdam, Netherlands)
Spiros Koulouzis (Informatics Institute, University of Amsterdam, Netherlands)
Adam S. Z. Belloum (Informatics Institute, University of Amsterdam, Netherlands)
In this paper we propose a novel method for auto-scaling data-centric work ow tasks. Scaling is achieved through a prediction mechanism where the input data load on each task within a work ow is used to compute the estimated task execution time. Through load prediction, the framework can take informed decisions on scaling multiple workflow tasks independently to improve overall throughput and reduce work ow bottlenecks. This method was implemented in theWS-VLAM work ow system and with an image analyses workflow we show that this technique achieves faster data processing rates and reduces overall workflow makespan.
Power-Aware Virtual Machine Scheduling on Clouds Using Active Cooling Control and DVFS
Daniel Guimaraes do Lago (Institute of Computing, State University of Campinas - UNICAMP, Brazil)
Edmundo R. M. Madeira (Institute of Computing, State University of Campinas - UNICAMP, Brazil)
Luiz Fernando Bittencourt (Institute of Computing, State University of Campinas - UNICAMP, Brazil)
This paper presents a virtual machine scheduling algorithm for Cloud Computing based on Green Computing concepts. The goal of this algorithm is the minimization of energy consumption in task executions in a cloud computing environment. This algorithm uses some features like shutdown of underutilized hosts, migration of loads of hosts that are operating below a certain threshold, and DVFS. It also applies the concept of active cooling control in order to minimize power consumption. The choice of which host will receive load is based on the concept of higher quality energy from the hosts, which is given by the ratio of MIPS consumed by the energy of each host. Results from simulation of this algorithm confronted with other surveyed algorithms have shown that it can improve the power consumption in clouds composed of heterogenous datacenters while being equivalent to the best algorithms in homogeneous datacenters.
Understanding Scheduling Implications for Scientific Applications in Clouds
Giacomo Mc Evoy (Distributed Scientific Computing (ComCiDis), National Laboratory for Scientific Computing - LNCC, Brazil)
Bruno Schulze (Distributed Scientific Computing (ComCiDis), National Laboratory for Scientific Computing - LNCC, Brazil)
This paper explores some of the ects that the paradigm of Cloud Computing has on schedulers when executing scientific applications. We present premises regarding to provisioning and architectural aspects of a Cloud infrastructure, that are not present in other environments, and which implications they may have on scheduling decisions in presence of relevant policies like improving performance. We also argue that using virtualization as a mechanism for workload consolidation in a multi-core environment has important performance consequences for e-science. We propose and test apreliminary workload classification, based on usage modes, that may improve early scheduling decisions as we research towards automatic deployment of scientific applications.
A Hybrid Fault Tolerance Scheme for EasyGrid MPI Applications
Jacques A. da Silva (Institute of Computing, Federal Fluminense University - UFF, Brazil)
Vinod E. F. Rebello (Institute of Computing, Federal Fluminense University - UFF, Brazil)
Writing applications capable of executing efficiently in distributed systems is extremely difficult and tedious for inexperienced users. The resources may be heterogeneous, nondedicated, and offered without any performance or availability guarantees. Systems capable of adapting the execution of an application to these characteristics are essential. The EasyGrid Application Management System (AMS) transforms cluster-based MPI applications into autonomic ones capable executing robustly and efficiently in distributed environments. This work describes a strategy to endow these autonomic MPI applications with the property of selfhealing and thus be capable of withstanding multiple simultaneous crash faults of processes and/or processors. The extremely low intrusion cost of the proposed hybrid solution might now facilitate acceptance of fault tolerance techniques in large scale high performance applications.
RvfcBOX: Multi-User Consistent File Sharing
Jean-Pierre Ramos (INESC ID / Technical University of Lisbon, Portugal)
Luis Veiga (INESC ID / Technical University of Lisbon, Portugal)
Paulo Ferreira (INESC ID / Technical University of Lisbon, Portugal)
The emerging of cloud file sharing systems has been motivated by real user needs for data sharing. There are many solutions providing such sharing support all having the common goal of being widely scalable while providing users with consistent shared data. However, oㄦing consistent data is at odds with scalability as it requires many messages and available network bandwith for file transfer. Network bandwidth can be minimized using several techniques such as compression, deduplication, delta encoding, etc. However, these approaches do not take into account that not all files must be fully consistent at all times for all users. In this paper we further increase the scalability of a cloud file sharing system, called vfcBOX, by taking into account the notion of users interest. This means that vfcBOX considers users' consistency needs regarding shared files, to avoid sending useless (or unnecessary) data through the network. As a matter of fact, some files do not need to be constantly propagated to all users, because some of them do not require such immediacy given the particular semantics of the shared data. vfcBOX uses not only deduplication techniques to minimize network usage but also a consistency model that takes into account the users' interests. The result is a scalable and encient cloud file sharing system that fulfills users needs regarding data sharing.