Minimizing impracticality of complex modeling via web service execution within a cloud-based platform

Abstract

Assessing the stability of stream banks is an important issue for those living within a floodplain, participating in activities such as hunting, rafting, and fishing, and for transporting water from one place to another safely. A stream’s stability is influenced by frequency and magnitude of streamflow, particle size, sediment transport, bed slope, and depth of the water (Allen, P.M., Arnold, J.G., Stinchcomb, 2012). Streams of all sizes adapt to streamflow regimes by continually altering channel properties such as channel’s shape and channel depth. As a result, assessing stream stability requires quantifying hydrologic factors that influence the stream’s shape and depth, especially under conditions of urbanization. Changes in climate and land use further complicate assessing a stream’s stability. This complication is due partly to humanity’s knowledge gaps or negligence in maintaining solutions of existing infrastructure (Vijay P. Singh, 2002). The most common cause of stream instability is the increase in surface runoff entering a stream due to urbanization. Engineers use models as a tool to help manage, address, or mitigate problems such as increasing surface runoff entering the streams. The primary goal of this research is to integrate an existing model (SWAT-DEG) as a web-based modeling tool to assess the hydrologic response to changes in climate and land use. Specific objectives of the study are: 1. To compliment an existing model by integrating the model into service oriented architecture. 2. To compliment an existing model by deploying a model utilizing web-services. 3. To explore the benefits of deploying an existing model as a web-based tool within a service-oriented architecture. Web services provide access to SWAT-DEG at any internet-accessible location, limits version control, independent of operating system, and eliminate desktop installations. Currently, SWAT-DEG is hosted by eRAMS.com’ (eRAMS) platform. eRAMS is an internet site that hosts and executes various models. The model is executed within a cloud environment to ensure the model can scale to increasing, simultaneous users. The cloud environment obtains scalable models by reducing the total time per request for a given model. The total time per request is defined as the time that elapses from when the user executes a function till the function is completed. For example, the total time per request spent for executing a model is the time spent between clicking run and the return of output from the model. To accomplish a reduction in total time per request, the execution of SWAT-DEG is parallelized within Cloud Services Innovation Platform (CSIP) (Lloyd, W, David, O, Lyon, J, Rojas, K.W., Ascough II, J.C., Green, T.R., Carlson, J.R., 2012) CSIP is a cloud infrastructure trying to implement modeling-as-a-service. Modeling-as-a-service is an attempt to provide the ability to scale various web-based models for large amounts of simultaneous users and/or expensive computational models. SWAT-DEG is broken up into two separate main functions within the source code, stochastic and deterministic. The stochastic part of SWAT-DEG executes the Monte Carlo (Ronald iii Christensen, Wesley Johnson, Adam Branscum, 2011; Yang, 2011) approach as the means to do uncertainty analysis. The deterministic approach is the same as the desktop version of SWAT-DEG. Scalability testing was applied to SWAT-DEG’s stochastic and deterministic services. The deterministic service showed clear benefits when executing multiple simultaneous users within the CSIP environment. However, a single user executing a deterministic service on the cloud is slower than a local computer. The stochastic service also saw a slight benefit from executing within the CSIP environment. The stochastic service did not receive a greater benefit due to a bottleneck in the output map reduction. Output map reduction is the process of taking multiple files and reducing them to a manageable size. SWAT-DEG processes this phase within one virtual machine (VM). A virtual machine is an online computer processor unit. Thus, by removing the limitation of executing map reduction within one VM, executing stochastic analysis within the cloud environment will result in far smaller total time per request.