In recent decades, disk storage capacity has increased spectacularly, but access speed to the information has not changed significantly. In light of the tremendous developments in semiconductor dynamic random-access memory (DRAM), it seems attractive to use it for online data, leaving disks for backup and archiving tasks. Initial attempts at this kind of work took place in the mid-1980s, for main memory databases.

This article makes a case for RAMCloud, a large-scale storage system that stores all of its information in the main memories of hundreds or thousands of commodity servers. RAMCloud promises 100 to 1,000 times better latency and throughput than disk-based systems. As the article states, “Though individual memories are volatile, RAMCloud can use replication and backup techniques to provide comparable data durability and availability.”

The RAMCloud concept explored in this article will accelerate the adoption of cloud computing, and “could have broad impact across the field of computing.” It is advantageous for sophisticated, intelligent algorithms, where access is random and unpredictable.