Alan Su

Adaptive computing on the Grid using AppLeS

Ensembles of distributed, heterogeneous resources, also known as computational grids, have emerged as critical platforms for high-performance and resource-intensive applications. Such platforms provide the potential for applications to aggregate enormous bandwidth, computational power, memory, secondary storage, and other resources during a single execution. However, achieving this performance potential in dynamic, heterogeneous environments is challenging. Recent experience with distributed applications indicates that adaptivity is fundamental to achieving application performance in dynamic grid environments. The AppLeS (Application Level Scheduling) project provides a methodology, application software, and software environments for adaptively scheduling and deploying applications in heterogeneous, multiuser grid environments. We discuss the AppLeS project and outline our findings.

Using Apples to Schedule Simple SARA on the Computational Grid

Computational Grids, composed of distributed and often heterogeneous computing resources, have become the platform of choice for many performance-challenged applications. Proof-of-concept implementations have demonstrated that both Grids and clustered environments have the potential to provide great performance benefits to distributed resource-intensive applications. However, at the present time, careful staging, scheduling, and/or reservation of resources is essential in order for applications to achieve performance in Grid environments. If Computational Grids and shared computational clusters are to achieve their full potential, it must be possible for users to achieve application performance at any given time, and when other users are present in the system. In this paper, we describe the initial development of an AppLeS (application-level scheduler) for the resource selection portion of the Synthetic Aperture Radar Atlas (SARA) application, developed at the Jet Propulsion Laboratory (JPL) and the San Diego Supercomputer Center (SDSC). We demonstrate the effectiveness of application scheduling for distributed data applications such as SARA by providing a performance-efficient strategy for retrieving SARA data files in everyday, multiple-user Grid environments.

Navigation and coordination primitives for multidimensional visual browsers

This paper describes extensions to the Tioga flight-simulator browsing protocol presented by Stonebraker et al. (1993a). These extensions allow users to navigate a multidimensional data space using sophisticated zooming capabilities. This design also allows users to move easily between different multidimensional spaces. Tunneling between different data spaces is shown to be a substantial generalization of hyperlinks in a hypermedia system. Finally, our design provides for the coordination of multiple browsers. This preserves context and allows users to explore multiple paths simultaneously.

Off-line scheduling of divisible requests on an heterogeneous collection of databanks

In this paper, we consider the problem of scheduling comparisons of motifs against biological databanks. We experimentally show that this problem lies in the divisible load framework with negligible communication costs. In this framework, we propose a polynomial-time algorithm to optimally solve the maximum weighted flow offline scheduling problem on unrelated machines. We also show how to optimally solve the maximum weighted flow off-line scheduling problem with preemption on unrelated machines.

Performance modeling for entity-level simulations

Advances across many fields of study are driving changes in the basic nature of scientific computing applications. Scientists have recognized a growing need to study phenomena by explicitly modeling interactions among individual entities, rather than by simply modeling approximate collective behavior. This entity-level approach has emerged as a promising new direction in a number of scientific fields. One of the challenges inhibiting the entity-level approach is the substantial resource requirements it entails. Unfortunately, such applications exhibit characteristics and behaviors which render traditional parallel computing techniques ineffective. Well-defined methodologies for achieving scalable performance on distributed computing platforms are needed. As an important first step, we present an abstract application model for entity-level applications, and we instantiate it for a case-study immunology application. Our experiments confirm that this model tracks application performance trends sufficiently well to study scheduling issues pertaining to entity-level applications. We identify a scalability problem inherent to the entity-level approach and use our model to quantify the potential performance improvements that remapping strategies may yield.

On the feasibility of running entity-level simulations on grid platforms

Scientists have long relied on abstract models to study phenomena that are too complex for direct observation and experimentation. As new scientific modeling methodologies emerge, new computing technologies must be developed. In this paper, we focus on entity-level modeling, a modeling approach that is gaining prevalence in many scientific fields. Although the principles of entity-level modeling are straightforward, entity-level simulations require a large amount of compute resource and grid platforms can meet such resource needs. Unfortunately, efficient large-scale distributed entity-level simulations have proven elusive, typically due to nondeterminism that renders classical distributed application deployment strategies ineffective. In this work, we propose a method for dynamically remapping application tasks to cope with this inherent nondeterminism. We evaluate the efficacy of this method in a simulated grid computing environment and discuss the feasibility of executing entity-level applications on grids.