GBRAMS

CLIMATOLOGY STUDIES USING BRAMS IN A

GRID COMPUTING ENVIRONMEN

Climatology simulations normally requires a high computational effort due to the large amount of data that must be processed. CPTEC/INPE is providing global model data for a 10 years climatology. Regional climatology can thus be obtained using an average over the ensemble, which is a set of BRAMS long range integration for each one of the 3 areas shown in Figure 1, with variations on initial conditions (member) by changing only the beginning time of integration, for 3 different initial dates.

The 3 areas and 3 members simulation represents 9 independent integrations. Therefore, these integrations can be executed in the same time, if there are computing resources avaliable. In order to the execution becomes more flexible, each N years integration, one area and one member is divided by year. This is made throught of the BRAMS checkpoint-restart mechanism (HISTORY mode) detailed in Figure 2.

Each N years integration is splitted in small executions (jobs) that, although they depend on each other, they can be executed in different computers as become avaliable.

STORAGE AND PROCESSING REQUIREMENTS

For 10 years of integration and 3 members, the total amount of boundary condition data necessary is nearly 58 Gb. The International Research Institute for Climate Prediction (IRI) standard strategy has been chosen for the time integration, that is performed on the whole period for each ensemble element. In Table 1 is presented the execution time in days of processing, running in a cluster with 16 Xeon 3.0 GHz processors, for 1 year of integration for each area, where BC is the boundary condition domain. Additionally, the total time for 10 years simulation and 3 members, is also estimed.

GRID COMPUTING

The Grid Computing concept was established by Foster [1] as a hardware and software environment with reliable, constant, and cheap access to computer facilities. In this work, besides the machine above described, the grid has also 2 CRAY/XD1 cluster, with 12 dual-core processors (Opteron 2.6GHz). As comparison, one year of integration in the SE area spent about 10% less time in these machines.

Several middlewares for Grid Computing are currently being developed. In this project, named GBRAMS, is made an on-going study of three solutions for Grid Computing, and of their effective usability to run meso-scale meteorological simulations: Globus [2], OurGrid [3] and OAR/CIGRI [1]. Each one of the research group involved in this work (LAC/INPE, CPTEC/INPE and II/UFRGS) offered a cluster in order to test a specific middleware in the grid.

GRID ARCHITECTURE

The interaction between the meteorologist and the grid is done throught a web portal (Figure 3), whose objective is to allow the creation of jobs and the results retrieval. In the Figure 4 is shown the grid architecture implemented. Each new job is inserted into a database, from where it is gotten later to execute.

A scheduler is in charge of getting the ready jobs from database and execute them in a grid computer. After execution, the results of post processed analysis are avaliable for viewing in the web portal (Figure 4). Graphics of monthly average temperature and precipitation are generated.

Temporal series of precipitation within each month of the year is also shown. So, based on these graphical outputs, a meteorologist can accept the result, allowing the execution of the next corresponding job. Otherwise, the job is can be executed again with new parameters values.

FINAL REMARKS

The proposed grid include the use of one of the 3 middlewares for 3 grid nodes (clusters at LAC/INPE, CPTEC/INPE and II/UFRGS). The presented performance figures of the BRAMS execution show that the climatology is feasible within reasonable execution times in the Xeon and Cray clusters.

It is expected that the GBRAMS project will validate the use of computational grids to allow higher number of simulated years and geographical extension in climatology studies. In addition, the increased computational power of grids may support simulations that couple climatology with other environmental models such as hydrological ones.

REFERENCES

[1] Foster I. and Kesselman C.: The Grid: Blueprint for a New Computing Infrastructure,Morgan Kaufmann, 1999.

[2] Foster I.: Globus Toolkit Version 4: Software for Service-Oriented Systems. IFIP International Conference on Network and Parallel Computing, Springer-Verlag LNCS 3779, pp 2-13, 2005.

[3] Cirne W. et al. : Building a User-Level Grid for BoT Applications. Book Chapter of High Performance Computing: Paradigm and Infrastructure. Laurence T. Yang, Minyi Guo, editors. John Wiley & Sons Inc., 2005.

[4] Capit N. et al. : A batch scheduler with high level components, Cluster computing and Grid 2005 (CCGrid05), 2005.

FAQ / e-mail