We present Apache Flink 2.0, an evolution of the popular stream processing system's architecture that decouples computation from state management. Flink 2.0 relies on a remote distributed file system (DFS) for primary state storage and uses local disks as a secondary cache, with state updates streamed continuously and directly to the DFS. To address the latency implications of remote storage, Flink 2.0 incorporates an asynchronous runtime execution model. Furthermore, Flink 2.0 introduces ForSt, a novel state store featuring a unified file system that enables faster and lightweight checkpointing, recovery, and reconfiguration with minimal intrusion to the existing Flink runtime architecture. Using a comprehensive set of Nexmark benchmarks and a large-scale stateful production workload, we evaluate Flink 2.0's large-state processing, checkpointing, and recovery mechanisms. Our results show significant performance improvements and reduced resource utilization compared to the baseline Flink 1.20 implementation. Specifically, we observe up to 94% reduction in checkpoint duration, up to 49× faster recovery after failures or a rescaling operation, and up to 50% cost savings.

The objective of this study is to measure the area of the winter wheat based on the multi-temporal middle resolution remote sensing images and the database of farmland parcel, and then compare the classification accuracy under the different amount of information. The study firstly analyzed the different crop samples in the support of field data, and then classified the images by the three different ways of NDVI threshold segmentation. The classification accuracies were compared in the three conditions of information amount: single-temporal image in November 19 in 2007, three-temporal image and three-temporal image with database of farmland parcel in Beijing in 2006. The results indicate that single-temporal image reveals the low classification accuracy. The multi-temporal images efficiently distinguish wheat from others plants. Moreover, the accuracy could be further improved if the database of farmland parcel was used to rule out the non-cultivated land.

Traditional methods always use NDVI to extract rice from remote sensing images, for example TM. However, little research focusses on introducing a moisture index into rice extraction. This paper introduced NDII and NDVI respectively to extract rice based on a TM image of a rice and ginkgo mixed area, and compared their ability in distinguishing rice and ginkgo from two aspects: position accuracy and regional accuracy. Research results show that: (1) NDII has the higher ability to distinguish rice and ginkgo; (2) rice extraction results from NDII are both better than results from NDVI, whether position accuracy or whole accuracy.