High Performance Land Surface Modeling with a Beowulf Cluster

1 We designed the Land Information System (LIS) to perform global land surface simulations at a resolution of 1-km or finer in real-time. Such an unprecedented scale and intensity in computing pose many challenges. In this article we will demonstrate some of our unique developments and implementations in high performance computing with a Beowulf Linux cluster, to meet these challenges and reach our performance goals. These contributions include a fault-tolerant job management system for the cluster, high-performance parallel I/O based on GrADS-DODS (GDS) servers with dynamic load-balancing and distributed data storage, and highly scalable data replication with peer-to-peer (P2P) technology. I