QFS (Quantcast File System) is an open-source distributed file system software package for large-scale MapReduce or other batch-processing workloads.
QFS is software that runs on a cluster of hundreds or thousands of commodity Linux servers and allows other software layers to interact with them as if they were one giant hard drive.
It has three components:
A chunk server runs on each machine that will host data, manages I/O to its hard drives, and monitors its activity and capacity.
A central process called the metaserver keeps the directory structure and maps of files to physical storage. It coordinates activities of all the chunk servers and monitors the overall health of the file system. For high performance it holds all its data in memory, writing checkpoints and transaction logs to disk for recovery.
A client component is the interface point that presents a file system API to other layers of the software. It makes requests of the metaserver to identify which chunk servers hold (or will hold) its data, then interacts with the chunk servers directly to read and write.
In a cluster of hundreds or thousands of machines, the odds are low that all will be running and reachable at any given moment, so fault tolerance is the central design challenge. QFS meets it with Reed-Solomon error correction. The form of Reed-Solomon encoding used in QFS stores redundant data in 9 places and is able to reconstruct the file from any 6 of these stripes.1 When it writes a file, it stripes it across by default nine physically different machines—six holding the data, three holding parity information. Any three of those can become unavailable; if any six remain readable, QFS can reconstruct the original data. The result is fault tolerance at a cost of a 50% expansion of data.
QFS is written in C++, operates within a fixed memory footprint, and uses direct I/O.
