GFS is one of those papers where every design decision sounds weird until you trace it back to the workload, and then it all clicks.

One master. Huge 64 MB chunks. No client-side data caching. Chunk locations rebuilt from heartbeats instead of persisted. Sounds wrong. Turns out to be exactly right for billions of files, hundreds of clients, and thousands of machines where something is always broken.

So I built it in Go to make sure I actually understood it. One file, no external packages. Master, chunkservers, client, leases, copy-on-write snapshots, pipelined data push, atomic record append — all of it, just the paper and a blank file.

Blog link: https://jitesh117.github.io/blog/google-file-system-in-go/