David Woollard

Scientific Software as Workflows: From Discovery to Distribution

Scientific workflows-models of computation that capture the orchestration of scientific codes to conduct in silico research-are gaining recognition as an attractive alternative to script-based orchestration. Even so, researchers developing scientific workflow technologies still face fundamental challenges, including developing the underlying science of scientific workflows. You can classify scientific-workflow environments according to three major phases of in silico research: discovery, production, and distribution. On the basis of this classification, scientists can make more-informed decisions regarding the adoption of particular workflow environments.

Injecting software architectural constraints into legacy scientific applications

While software architectures have been shown to aid developers in maintenance, reuse, and evolution as well as many other software engineering tasks, there is little language-level support for these architectural concepts in scientific programming languages such as Fortran and C. Because many existing scientific codes are written in legacy languages, it is difficult to integrate them into architected software systems. By wrapping scientific codes in architecturally-aware interfaces, we are able to componentize legacy programs, integrating them into systems built with first-class architectural elements while meeting performance and throughput requirements of scientific codes.

A Science Data System approach for the SMAP mission

Though Science Data System (SDS) development has not traditionally been part of the mission concept phase, lessons learned and study of past Earth science missions indicate that SDS functionality can greatly benefit algorithm developers in all mission phases. We have proposed a SDS approach for the SMAP Mission that incorporates early support for an algorithm testbed, allowing scientists to develop codes and seamlessly integrate them into the operational SDS. This approach will greatly reduce both the costs and risks involved in algorithm transitioning and SDS development.

Exploiting connector knowledge to efficiently disseminate highly voluminous data sets

Ever-growing amounts of data that must be distributed from data providers to consumers across the world necessitate a greater understanding of the software architectural implications of choosing data movement technologies. Currently, this understanding is mired in the minds of software architects who have been there before, and who rely on past intuition and choices, failing to properly document their rationale and context. In this paper we describe a software architecture-based decision making framework called DISCO for selecting data movement technologies, or software connectors. DISCO effectively captures (traditionally undocumented) insight, observation and ultimately architectural knowledge about the connectors, demonstrating the effectiveness of using such information to accurately encode the connector selection decision making process

Kadre: domain-specific architectural recovery for scientific software systems

Scientists today conduct new research via software-based experimentation and validation in a host of disciplines. Scientific software represents a significant investment due to its complexity and longevity yet there is little reuse of scientific software beyond small libraries which increases development and maintenance costs. To alleviate this disconnect, we have developed KADRE, a domain-specific architecture recovery approach and toolset to aid automatic and accurate identification of workflow components in existing scientific software. KADRE improves upon state of the art general cluster techniques, helping to promote component-based reuse within the domain.

A Distributed Computing Infrastructure for the Evaluation of Climate Models Using NASA Observational Data

We describe and motivate an emerging distributed data analysis environment for the empirical evaluation of climate models. This work has several components, including: model scoring, initialization and parameterization, all of which require massive amounts of NASA observational data. Though the effort is in its nascence, there has been recent success in partnering with the DOE Earth System Grid to unify data from NASA’s AIRS mission and model outputs available from the Program for Climate Model Diagnosis and Intercomparison (PCMDI).

An architectural style for high-performance asymmetrical parallel computations

Researchers with deep knowledge of scientific domains are becoming more interested in developing highly-adaptive and irregular (asymmetrical) parallel computations, leading to development challenges for both delivery of data for computation and mapping of processes to physical resources. Using software engineering principles, we have developed a new communications protocol and architectural style for asymmetrical parallel computations called ADaPT.Utilizing the support of architecturally-aware middleware, we show that ADaPT provides a more efficient solution in terms of message passing and load balancing than asymmetrical parallel computations using collective calls in the Message-Passing Interface (MPI) or more advanced frameworks implementing explicit load-balancing policies. Additionally, developers using ADaPT gain significant windfall from good practices in software engineering, including implementation-level support of architectural artifacts and separation of computational loci from communication protocols.