Distributed file system

Design a distributed file system for a data/analytics platform (GFS/HDFS-style) that backs large sequential reads. Constraints: files are mostly huge (GBs to TBs), write-once/append-mostly with concurrent appends, throughput matters more than latency, hundreds of PB across thousands of commodity nodes, and the system must tolerate frequent node failures. Cover the namespace, chunking, and replication.