> <@_discord_417857235079790592:mailscript.com> various_accounts: yes it's from the same team
> 
> odisseus: it really doesn't take much data to make go-ipfs start to fall apart. Once you start storing 1TB of data you literally can't garbage collect your node. It means the only practical way of removing data is by nuking your data directory and copying back the data you want.
> 
> neoflux: no you don't need to reach Netflix scale before IPFS becomes unusable at scale. The moment you start storing terabytes of data, go-ipfs becomes absolutely impractical

We've solved the ipfs GC problem by implementing our own GC for Peergos. It's fully concurrent (doesn't block any user actions), and operates directly on the block store. This is made possible because of us storing the pin set outside of ipfs, as well as a few other features of our architecture.