Cagra is a distributed storage system optimized for distributed rendering. Enterprise-class distributed rendering requires computers with considerable storage space and computational power. An single frame in an average 3D CG movie requires approximately 1 GB to 20 GB of input data and takes 3 to 10h to render. The large-scale computations in rendering are carried out by specialized clusters, or render farms. However, a centralized storage system is used for storing input data. Cagra is designed as a lightweight distributed key-value store optimized for use in a rendering environment. In this paper, we discuss the design and implementation of Cagra and its various features such as high dependability, dynamic node insertion, and zero configuration. We also discuss details of its optimization for rendering in order to increase the rendering throughput.