Chapter 7 Implementation of DECGrid and MODO Services
7.5 Services implementation
7.5.3 Search strategy selection
It is important to use the appropriate search algorithm for every model. This can be done by providing the characteristics of the model to a query form. The properties could include number of objectives, types of objectives, number of constraints and number of variables and variable types (discrete or continuous). The query will match these properties to appropriate algorithms. Multi-objective design optimisation (MODO) is not aimed at finding a single optimum solution, but multiple optimum solutions known as Pareto solutions. Each of these solutions is known as Pareto optimum (Nebro et al., 2007). The search strategy in this case is to find the Pareto front (graph of points joining each Pareto optimum solution). MODO uses three main search strategies namely enumerative, deterministic and stochastic strategies (Luna et al., 2006). Enumerative and deterministic have finite population search space, though they can be computationally and data intensive after many iterations of computations.
Stochastic uses dynamic search population space. This research only provided NSGA- II algorithm which is a stochastic algorithm. However, the Globus search browser interface is configured to accommodate enumerative and deterministic algorithms in future research.
Using grid computing technology, a search strategy selection service based on Globus Web Services Monitoring and Discovery System (WS MDS) is developed to suit enumerative and deterministic strategies on one hand and stochastic strategy on the other hand. MDS is used to monitor and discover MODO resources such as optimisation algorithms that fall under the different search categories within a virtual organisation (VO). The Index Service collects monitoring and discovery information from optimisation resources and publishes it in one location for requestors to access. An intuitive and user-friendly web interface known as service group provides requestors, in this case optimisation engineers to view and query grid resources, in this case optimisation resources. This is done using the Open Grid Service Architecture (OGSA) browser graphical user interface. This browser allows users to query resources that are registered. The browser has pre-configured (fixed) and open- ended (dynamic) query capability. Pre-configured query searches for particular resources while dynamic query looks for resources that have dynamic or varying properties. The first stage of using the search strategy selection is to run the WebMDS and the browser appears with options to view all resources or particular resources using special query forms. There are basically four default query forms on the browser. Two are preconfigured queries and so are suitable for enumerative and deterministic search strategies while the other two are configured to be used for stochastic search strategy. Figure 7.7 shows the diagram of the search strategy selection query form. Although in this research, only NSGA-II is used for the 3 case studies, the search strategy selection (SSS) functionality is aimed at making the system robust enough to accommodate other algorithms.
Preconfigured queries are aimed at specific optimisation processes. The output is usually predicted based on experience. The search focuses on a determined direction. This strategy is suitable for enumerative and deterministic search strategies. Open Ended Optimisation Queries adopts strategy that searches optimisation resources in many sources at different grid nodes. These are suitable for stochastic search strategy.
The clients (optimisation engineers) submit queries to optimisation service registry (see Figures 5.16 and 5.25) to search for a search strategy. Clients intending to use stochastic strategy submit queries to view all the resources published through a single source or many sources. The service group also displays information of resources and users can submit queries through it.
Search Strategy Selection Query Form
Client
Preconfigured Queries
Preconfigured Queries & Transforms Open-Ended Optimisation Resource Query Open-Ended Optimisation Resource Request Information Sources Single Source Optimisation Resource Query Many Sources Optimisation Query Optimisation Service Registry Optimisation Service Group
Submit Query Submit Query Submit Query Submit Query Submit Query
Client Client Client Client
Enumerative/ Deterministic Enumerative/ Deterministic Stochastic Stochastic Resource Properties: EP URL, EP Key Name, EP Key Value, RP Namespace
Figure 7.7: First stage of search strategy selection optimisation resource query process (end point =EP; resource property=RP; uniform resource locator=URL)
Figure 7.8 shows the search strategy selection procedure. Optimisation resources are first published by providers using the Globus RegistrationAggregatorSource component of the Aggregator Source service and then requestors (optimisation engineers) use the web service group to view available optimisation resources and determine which ones are suitable for their jobs. After determining the resources to be used, the Globus SubscriptionAggregatorSource is triggered for requestors to
subscribe resources. The optimisation expert could invoke the
to other resources such as Globus Access and Secondary Storage (GASS) which may hold the design parameters and GridFTP (Grid File Transfer Protocol) which is used to transfer the input files. Alternatively, there are parameter and optimisation services that perform input parameter and optimisation tasks separately. The main task of the search strategy selection is to enable easy selection of a search strategy that best suits the optimisation problem in question. Each algorithm is identified by its properties (classical or stochastic), endpoint (EP), EP key, EP value or other resource properties (RP). End point (EP) is the identification tags for registered resources and queries are made by entering EPs in the browser to search for the resource. The optimisation can be split into quantitative (QT) and qualitative (QL) components which are handled by the QTand QLservices respectively.
Register Optimisation Resources
Search Strategy Selection Query Form Subscribe Optimisation Resources QT QL Storage Optimisation Output Good? Yes No Stop By providers using By requestors using
Globus Access Secondary Storage Execute Optimisation Resources By requestors using RegistrationAggregatorSource SubscriptionAggregatorSource ExecutionAggregatorSource Math Model Building
Search Strategy Selection
The search strategies in the grid are accomplished by understanding the way services are provided using standard XML (Extensible Mark-up Language) schema. This is one of the objectives of the research (to understand grid environment requirements for MODO application services). This research used the Globus toolkit service schema (similar to Figure 7.4) to implement all the MODO services provided in DECGrid.
The advantage of this concept is that it allows optimisation engineers to search for the optimisation algorithms that best suit their requirements and the flexibility to try it over again with different algorithms at the same time. Search strategy selection interface allows viewing and querying of registered resources. The aggregator source (see section 5.5.3) accepts registered resources and allows subscription and execution of the services.