Yaakoub El-Khamra

Autonomic management of application workflows on hybrid computing infrastructure

In this paper, we present a programming and runtime framework that enables the autonomic management of complex application workflows on hybrid computing infrastructures. The framework is designed to address system and application heterogeneity and dynamics to ensure that application objectives and constraints are satisfied. The need for such autonomic system and application management is becoming critical as computing infrastructures become increasingly heterogeneous, integrating different classes of resources from high-end HPC systems to commodity clusters and clouds. For example, the framework presented in this paper can be used to provision the appropriate mix of resources based on application requirements and constraints. The framework also monitors the system/application state and adapts the application and/or resources to respond to changing requirements or environment. To demonstrate the operation of the framework and to evaluate its ability, we employ a workflow used to characterize an oil reservoir executing on a hybrid infrastructure composed of TeraGrid nodes and Amazon EC2 instances of various types. Specifically, we show how different applications objectives such as acceleration, conservation and resilience can be effectively achieved while satisfying deadline and budget constraints, using an appropriate mix of dynamically provisioned resources. Our evaluations also demonstrate that public clouds can be used to complement and reinforce the scheduling and usage of traditional high performance computing infrastructure.

Exploring the Performance Fluctuations of HPC Workloads on Clouds

Clouds enable novel execution modes often supported by advanced capabilities such as autonomic schedulers. These capabilities are predicated upon an accurate estimation and calculation of runtimes on a given infrastructure. Using a well understood high-performance computing workload, we find strong fluctuations from the mean performance on EC2 and Eucalyptus-based cloud systems. Our analysis eliminates variations in IO and computational times as possible causes, we find that variations in communication times account for the bulk of the experiment-to-experiment fluctuations of the performance.

An Autonomic Approach to Integrated HPC Grid and Cloud Usage

Clouds are rapidly joining high-performance Grids as viable computational platforms for scientific exploration and discovery, and it is clear that production computational infrastructures will integrate both these paradigms in the near future. As a result, understanding usage modes that are meaningful in such a hybrid infrastructure is critical. For example, there are interesting application workflows that can benefit from such hybrid usage modes to, per- haps, reduce times to solutions, reduce costs (in terms of currency or resource allocation), or handle unexpected runtime situations (e.g., unexpected delays in scheduling queues or unexpected failures). The primary goal of this paper is to experimentally investigate, from an applications perspective, how autonomics can enable interesting usage modes and scenarios for integrating HPC Grid and Clouds. Specifically, we used a reservoir characterization application workflow, based on Ensemble Kalman Filters (EnKF) for history matching, and the CometCloud autonomic Cloud engine on a hybrid platform consisting of the TeraGrid and Amazon EC2, to investigate 3 usage modes (or autonomic objectives) – acceleration, conservation and resilience.

Distributed Application Runtime Environment (DARE): A Standards-based Middleware Framework for Science-Gateways

Gateways have been able to provide efficient and simplified access to distributed and high-performance computing resources. Gateways have been shown to support many common and advanced requirements, as well as proving successful as a shared access mode to production cyberinfrastructure such as the TG/XSEDE. There are two primary challenges in the design of effective and broadly-usable gateways: the first revolves around the creation of interfaces that catpure existing and future usage modes so as to support desired scientific investigation. The second challenge and the focus of this paper, is concerned about the requirement to integrate the user-interfaces with computational resources and specialized cyberinfrastructure in an interoperable, extensible and scalable fashion. Currently, there does not exist a commonly usable middleware to that enables seamless integration of different gateways to a range of distributed and high-performance infrastructures. The development of multiple similar gateways that can work over a range of production cyberinfrastructures, usage modes and application requirements is not scalable without a effective and extensible middleware. Some of the challenges that make using production cyberinfrastructure as a collective resource difficult are also responsible for the absence of middleware that enables multiple gateways to utilize the collective capabilities. We introduce the SAGA-based, Distributed Application Runtime Environment (DARE) framework, using which gateways that seamlessly and effectively utilize scalable distributed infrastructure can be built. We discuss the architecture of DARE-based gateways, and show using several different prototypes—DARE-HTHP, DARE-NGS, how gateways can be constructed by utilizing the DARE middleware framework.

Exploring application and infrastructure adaptation on hybrid grid-cloud infrastructure

Clouds are emerging as an important class of distributed computational resources and are quickly becoming an integral part of production computational infrastructures. An important but oft-neglected question is, what new applications and application capabilities can be supported by clouds as part of a hybrid computational platform? In this paper we use the ensemble Kalman-filter based dynamic application workflow and investigate how clouds can be effectively used as an accelerator to address changing computational requirements as well as changing Quality of Service constraints (e.g., deadlines). Furthermore, we explore how application and system-level adaptivity can be used to improve application performance and achieve a more effective utilization of the hybrid platform. Specifically, we adapt the ensemble Kalman-filter based application formulation (serial versus parallel, different solvers etc.) so as to execute efficiently on a range of different infrastructure (from High Performance Computing grids to clouds that support single core and many-core virtual machines). Our results show that there are performance advantages to be had by supporting application and infrastructure level adaptivity. In general, we find that grid-cloud infrastructure can support novel usage modes, such as deadline-driven scheduling, for applications with tunable characteristics that can adapt to varying resource types.

Developing autonomic distributed scientific applications: a case study from history matching using ensemblekalman-filters

The development of a simple effective distributed applications that can utilize multiple distributed resources remains challenging. Therefore, not surprisingly, it is difficult to implement advanced application characteristics - such as autonomic behaviour for distributed applications. Notwithstanding, there exist a large class of applications which could benefit immensely with support for autonomic properties and behaviour. For example, many applications have irregular and highly variable resource requirements which are very difficult to predict in advance. As a consequence of irregular execution characteristics, dynamic resource requirements are difficult to predict a priori thus rendering static resource mapping techniques such as work flows ineffective; in general the resource utilization problem can be addressed more efficiently using autonomic approaches. This paper discusses the design and development of a prototype framework that can support many of the requirements of Autonomic applications that desire to use Computational Grids. We provide here an initial description of the features and the architecture of the Lazarus framework developed using SAGA, integrate it with an Ensemble Kalman Filter application, and demonstrate the advantages - performance and lower development cost, of the framework. As proof of concept we deploy Lazarus on several different machines on the TeraGrid, and show the effective utilization of several heterogeneous resources and distinct performance enhancements that autonomics provides. Careful analysis provides insight into the primary reason underlying the performance improvements, namely a late-binding and an optimal choice of the configuration of resources selected.

Performance evaluation of R with Intel Xeon Phi coprocessor

Over the years, R has been adopted as a major data analysis and mining tool in many domain fields. As Big Data overwhelms those fields, the computational needs and workload of existing R solutions increases significantly. With recent hardware and software developments, it is possible to enable massive parallelism with existing R solutions with little to no modification. In this paper, we evaluated approaches to speed up R computations with the utilization of the Intel Math Kernel Library and automatic offloading to Intel Xeon Phi SE10P Co-processor. The testing workload includes a popular R benchmark and a practical application in health informatics. There are up to five times speedup gains from using MKL with a 16 cores without modification to the existing code for certain computing tasks. Offloading to Phi co-processor further improves the performance. The performance gains through parallelization increases as the data size increases, a promising result for adopting R for big data problem in the future.

Developing Scientific Applications with Loosely-Coupled Sub-tasks

The Simple API for Grid Applications (SAGA) can be used to develop a range of applications which are in turn composed of multiple sub-tasks. In particular SAGA is an effective tool for coordinating and orchestrating the many sub-tasks of such applications, whilst keeping the application agnostic to the details of the infrastructure used. Although developed primarily in the context of distributed applications, SAGA provides an equally valid approach for applications with many sub-tasks on single high-end supercomputers, such as emerging peta-scale computers. Specifically, in this paper we describe how SAGA has been used to develop applications from two types of applications: the first with loosely-coupled homogeneous sub-tasks and, applications with loosely-coupled heterogeneous sub-tasks. We also analyse and contrast the coupling and scheduling requirements of the sub-tasks for these two applications. We find that applications with multiple sub-tasks often have dynamic characteristics, and thus require support for both infrastructure-independent programming models and agile execution models. Hence attention must be paid to the practical deployment challenges along with the theoretical advances in the development of infrastructure-independent applications.

A Parallel High-Order Discontinuous Galerkin Shallow Water Model

The depth-integrated shallow water equations are frequently used for simulating geophysical flows, such as storm-surges, tsunamis and river flooding. In this paper a parallel shallow water solver using an unstructured high-order discontinuous Galerkin method is presented. The spatial discretization of the model is based on the Nektar++ spectral/hp library and the model is numerically shown to exhibit the expected exponential convergence. The parallelism of the model has been achieved within the Cactus Framework. The model has so far been executed successfully on up to 128 cores and it is shown that both weak and strong scaling are largely independent of the spatial order of the scheme. Results are also presented for the wave flume interaction with five upright cylinders.

Empowering R with High Performance Computing Resources for Big Data Analytics

The software package R is a free, powerful, open source software package with extensive statistical computing and graphics capabilities. Due to its high-level expressiveness and multitude of domain specific packages, R has become the lingua franca for many areas of data analysis, drawing power from its high-level expressiveness and its multitude of domain specific, community-developed packages. While R is clearly a “high productivity” language, it has not necessarily been a “high performance” language. Challenges still remain in developing methods to effectively scale R to the power of supercomputers, and in deploying support and enabling access for the end users. In this chapter, we focus on approaches that are available in R that can adopt high performance computing resources for providing solutions to Big Data problems. Here we first present an overview of current approaches and support in R that can enable parallel and distributed computations in order to improve computation scalability and performance. We categorize those approaches into two on the basis of the hardware requirement: single-node parallelism that requires multiple processing cores within a computer system and multi-node parallelism that requires access to computing cluster. We present a detail study on performance benefit of using Intel® Xeon Phi coprocessors (Xeon Phi) with R for improved performance in the case of single-node parallelism. The performance is also compared with using general-purpose graphic processing unit through HiPLAR package and other parallel packages enabling multi-node parallelism including SNOW and pbdR. The results show advantages and limitations of those approaches. We further provide two use cases to demonstrate parallel computations with R in practice. We also discuss a list of challenges in improving R performance for the end users. Nevertheless, the chapter shows the potential benefits of exploiting high performance computing with R and recommendations for end users of applying R to big data problems.