A RAMCloud Storage System based on HDFS: Architecture, implementation and evaluation

Few cloud storage systems can handle random read accesses efficiently. In this paper, we present a RAMCloud Storage System, RCSS, to enable efficient random read accesses in cloud environments. Based on the Hadoop Distributed File System (HDFS), RCSS integrates the available memory resources in an HDFS cluster to form a cloud storage system, which backs up all data on HDFS-managed disks, and fetches data from disks into memory for handy accesses when files are opened for read or specified by users for memory storage. We extend the storage capacity of RCSS to that of the substrate disk-based HDFS by multiplexing all the available memory resources. Furthermore, RCSS supports MapReduce, which is a popular cloud computing paradigm. By serving data from memory instead of disks, RCSS can yield high random I/O performance with low latency and high throughput, and can achieve good availability and scalability as HDFS.

► An HDFS-based RAMCloud Storage System (RCSS) is proposed to improve random I/O performance for HDFS-based applications.
► RCSS is designed and implemented in Java, and is now available for application deployment.
► An performance evaluation on RCSS is conducted, and performance comparisons among RCSS, RAM-HDFS and HDFS are provided.