- Those popular generic simulation models have been implemented world-wide in a large number of water systems and incorporate most of the desirable attributes of a simulation model. - The DSS is specifically developed to meet the system management requirements to satisfy the growing demands in multi-reservoir systems under water scarcity conditions, as frequently, happen in the Mediterranean regions. - Sechi and Sulis (2009a) have recently developed a full integration of the simulation module WARGI- SIM and the linear optimization module WARGI-OPT in the DSS. - A common disadvantage of the traditional modeling approach is the large number of system simulations required to achieve acceptable levels of confidence treating data uncertainties in the model. - A grid is both hardware and software services, in the field of the advanced calculation and of the nets data transmissions to high speed network that allows to various geographic sites to share the own resources in dynamics and intelligent way and allowing the transparent, secure, controlled access by multiple users for public or private scientific and technologic research. - The highest layer of the grid architecture is the application layer, which includes applications in science, engineering, business, finance and more, as well as portals and development toolkits to support the applications. - A Virtual Organization (VO) represents a fundamental concept of Grid Computing technology: it is a group of grid users with similar interests and requirements who are able to work collaboratively with other members of the group and/or share resources (data, software, expertise, CPU, storage space, etc.) regardless of geographical location. - Grid computing is the IT technology enabling worldwide scientific projects, such as the Large Hadron Collider (LHC) at CERN, and powering global efforts to combat climate change, discover new medicines, map the skies, reconstruct the sound of ancient instruments, covers some aspects of the preservation and the fruition of cultural heritages and so on. - 7/2007) on the porting of the WARGI-DSS to the Italian National GRID. - Specifically, this paper focuses on the possible use of the combined simulation and optimization approach in the WARGI-DSS within the GRID environment for the definition of drought mitigation measures based on a large number of possible future system evolutions. - In Section 2, an overview of the development and features of WARGI-DSS is presented. - Section 3 describes the implementation of the GRID approach to satisfy the requirements of massive simulation- optimization runs. - Sections 4 and 5 show a practical application to a complex water system in the Mediterranean area, and conclusions and perspective of future works.. - An overview of the WARGI-DSS 2.1 The WARGI-DSS structure. - Moreover, the software modularity allows easy coding changes and the addition of new objects and features in the system diagram.. - The SDI module handles the values definition of the main parameters and the creation and possible modification of system elements. - Moreover, water quality optimization considering synthetic quality indexes for water sources and demands is implemented in the WARGI-QUAL module (Sechi and Sulis, 2007b, 2009b).. - As usual, the preliminary requirement of WARGI-SIM is to represent in the model all the features the user thinks are important with respect to the objectives of the study. - This enables WARGI-SIM to identify the technical and economic constraints of the system being modelled.. - In the combined approach, WARGI-SIM processes the flows configuration provided by WARGI-OPT when foreseeing for many different hydrologic scenarios. - This mixed optimization-simulation approach has to manage a large number of generated hydrologic scenarios and operating rules that might be tested in the simulation model and the GRID computing is very attractive to face this complex computation problem.. - GRIDs appear to be promising in the optimization and simulation of complex systems but few applications have focused on the analysis of multi-reservoir systems (Sulis, 2009).. - Each graphic element of the graph is denominated GOB (Graphical OBject) and is identified by both the type of GOB and the “identifying value” called Tk-id. - WARGI uses keys of the char* type to quickly localize its entry in the hash table. - Such a key is associated to a value which is a pointer to the initial position of the hash table containing the information on the GOB attributes. - initialized by the information contained in the TCL variables. - these allow the possibility of accessing information on the attributes of the GOB directly by the execution of script TCL, or directly through the calling of function C or C. - The communication between variables TCL and C/C++ is guaranteed by the mechanism called “variable linking” which allows the value of the TCL variables to be associated to the C/C++ variables and vice versa. - In the graph of the water system the correspondence node-arcs is represented by a matrix of adjacency. - The information on the global variables (year-length simulation and optimization horizons, number of periods per year, optimization-simulation linking period) is stored in the TCL and C/C++ global variables. - The information of the optimization model can be saved in a file with idr extension. - In such a case, the graph is showed in the main window of the graphical user interface (Figure 3). - Main windows in the GUI.. - A specific procedure creates and sets the graphic object of the GUI: the canvas, the palette, the menu bar and the relative drop down menus, the scroll and the state bars. - The canvas allows user to create, locate and connect the components of the water system as shown in Figure 3. - The core of the canvas is the graph of nodes and arcs (GOBs). - The graph reflects the spatial relationships between the elements of the water system. - Each GOB must be associated with a hash table consisting of as many fields as the number of attribute in the GOB. - The user can activate the GOB, open a window associated with that GOB and populate the hash table with the attributes of the GOB. - The window of the reservoir GOB is shown in Figure 4 and the corresponding hash table must contain numerical information on the physical, hydrologic and management attributes of the reservoir. - Reservoir window in the GUI.. - The menu bar provides access to the most important features of the DSS. - Time window setting in the GUI.. - To perform the analysis of a water system, the user has to input the values of the global variables in the Time Menu (Figure 5):. - The WARGI-OPT module forecasts the system evolution on the time horizon Δ at each synchronization period τ i based on the state indicators of the system [I] and a user-selected future hydrological synthetic scenario [b g ] (Figure 6). - they are decision variables of the source GOBs. - The hydrological scenarios [b g ] are attributes of the reservoir GOBs.. - The linear optimization model can be therefore written in the following form:. - In the WARGI-SIM module, preferences and priorities [v] are attributes of sources and demands GOBs, respectively. - τ τ τ τ = (5) The decision variables of water allocation [X t ] in the system are the solution of a minimum. - In the case of water scarcities more severe than those forecasted by WARGI-OPT, the preemptive measures [z τ ] do not make it possible to overcome the water scarcity, and WARGI-SIM introduces further restriction measures [s t ] in a reactive approach. - These reactive actions are defined following the state indicators of the system [I t. - The goal of this mixed optimization-simulation approach is to define the best combination of drought mitigation measures that minimizes the economic impact of drought in the water. - The economic response function R is the sum of the costs associated with the construction of new works in the system ([C γ. - [CP D ] and [C NPD ] are associated with the drought mitigation measures in the proactive and reactive approaches, respectively. - In the proposed GRID approach (Figure 7), WARGI-SIM considers together n-sets of decision variables [x i , x j ] (5) from the solution of n linear optimization models in the GRID environment. - As expected, the user can get a significant reduction of computational time saving by applying the proposed GRID approach instead of the traditional local approach. - Details of the porting to the GRID environment of this integrated approach are presented in this section.. - Link between WARGI-OPT and WARGI-SIM in the proposed GRID environment.. - 3.1 Porting of the mixed optimization-simulation approach in WARGI-DSS. - The middleware chosen for the implementation of the optimization-simulation approach is gLite. - In fact with gLite users can exploit directly resources, at low level, or let the so called central services the task of selecting, among those available, the most appropriate resource for the execution of the computational task. - The request for the job execution is the provision to the WMS of the file containing the job description. - as outcome of the request, a job identifier is returned. - The advantage of this approach is that in case of big input files, they are transferred once to the resource, providing so an optimization of the bandwidth usage. - In this case, the job forming the collection have the common part made by the solver, while the sub-jobs differs each other for the different instance of the optimization to be solved.. - While this approach is simple, and yet very powerful, allowing any kind of interactions, has the limit of the considerable computational overhead taken by the command line calls made from inside the application. - For this reason, the approach with API is sometimes preferable, because it carries better performances and a cleaner design of the application interacting with the grid environment. - So, paying the price of a longer develop time, because application source code needs to be adapted, API’s provide better performances, and a cleaner design of the grid application. - The effort needed for the inclusion of API libraries is reduced by the choice of the embedding of only the needed functions, rather than the whole set of gLite API. - the location of user credential, the URL of the WMS service and the JDL which describes the submitting job. - So, if they don’t exist or they are unknown, all these parameters have to be generated before the execution of an instance of the application, but then can be then reused for further instances.. - The Logging and Bookkeeping service (LB) is a central registry, where are logged all the steps of the job life cycle, for instance when it has been assigned to a resource, or when it’s being executed or if some errors happens. - When performing the next query, the application will find the Done status, and, by means of Job Output retrieval, proceed to the download of the Output Sandbox, which contains the solutions as computed from the solver.. - The Flumendosa-Campidano water system (Figure 8) is located in the South Sardinia, Italy.. - Table 1 shows the main statistical properties of the available hydrological series at the sites of interest.. - Table 2 shows proactive and reactive measures currently applied and planned in the Flumendosa-Campidano water system. - The WARGI DSS has been applied to the system to provide the water authority of the Flumendosa-Campidano with a first estimate of the relationship between hydrologic uncertain data and drought mitigation measures.. - The water authority often need to identify the sensitivity of the system economic performance associated with changes in hydrology and to quantify the consequences of alternative hydrological assumptions about the future. - Consequently, the analysis of the system should be done with a wider range than, as usual, the only “best guessing”. - Figure 9 summarizes the results of this ex-post economic analysis showing the values of response function R of the Flumendosa-Campidano system for each of the seven selected hydrological series. - In the GRID approach those 7 linear programming models related to the selected hydrological series 2 year length, are solved at each synchronization period τ i by LpSolve running in different Computing Elements. - Over the 54 time period of Flumendosa-Campidano system analysis, the value of the economic function R equals 1.39 million Euros per year. - As shown in Figure 10, the computation time in the local approach vastly increased and the software simulated the system in about 21 min when the seven selected hydrological series were considered in the optimization phase. - At that time of the presented research, there were 10 Intel Pentium Dual CPU running Scientific Linux SLC in the GRID authorized for a given job. - The total Grid computation time is defined by the Grid latency (the time needed to submit a complete set of jobs, and collect a complete set of results), the waiting time in case a job is delayed in the Grid queues and the running time on the nodes. - The total computation time was reduced more than 60% in the case of seven hydrological series.. - Response function trend of the system for different hydrological series.. - If we keep constant the number of operation totally executed by a collection, the global computation time decreases proportionally with the number of jobs executed in parallel in the GRID nodes (until a certain threshold is passed, as we will see). - The times needed for the job submission and matchmaking don’t need to be accounted, because in the job collection implementation, there is always one submission and one matchmaking, independently from the number of jobs actually forming the collection. - This quantity can be computed as the time that each job waits in the resource queue, plus its real running time. - Since it’s not known the number of resources in advance, and their availability status, a rule of thumb is that collection size should not exceed half of the free CPUs available, at the submission, for the testbed. - Otherwise, it’s likely that one or more of the collection jobs will be assigned to a slow resource, degrading the global execution time. - However in the authors’ view, there are still some hindrances for an extensive use in the water system field. - A larger number of hydrological series used in the optimization phase to account for all the relevant statistics of drought events and different weights for the correspondent LP models should be of interest. - Future research in the area of GRID approach for this combined optimization-simulation model also include its application in a more realistic GRID environment to estimate how many hydrological series is convenient to add before the performance of the GRID approach will certainly decrease.. - A comparison, evaluation, and identification of future needs, In: Environmental Modelling, Software and Decision Support (3): The State of the Art and New Perspective, Jakeman, Voinov, Rizzoli &. - Proceedings of the iEMSs Third Biennial Meeting: "Summit on Environmental Modelling. - In: The Operation of Multiple Reservoir Systems - IIASA Collaborative Proceedings Series CP-82-S3, Kaczmarek &. - Tools for water management: A view of the future