Scaling Namespace Operations with Giraffa File System

HDFS clusters rely on a single NameNode, the master, as its metadata service. Single master design of HDFS is known to be a limiting factor for potential growth of the file system in its size and performance. Project Giraffa replaces the single master of HDFS with a dynamically distributed namespace service, thus overcoming scalability limits of HDFS while remaining fully compatible with it. We focus on the performance of namespace operations and present a benchmark that demonstrates that Giraffa can linearly scale the throughput of metadata operation by simply adding more servers to store the file-system namespace.