The scaling issue with IPFS is well documented, and i wish it would just be one thing.

off the top of my head

1. A healthy IPFS repo is simply not maintainable after a certain size. From my experience, anything above 2 terabytes (per repo) is not practical. You can't run garbage collections, they become too slow, and even stat operations can hang.

2. Normally one would want a 10 RAID for a video platform's catalog. Makes sense, but given the nature of IPFS, a RAID is simply too slow. It chokes out the server easily with mid traffic. Yes, it keeps things safe, redundancy, but at the cost of speed.

3. The IPFS DHT is mainly noise. Lookups on IPFS, if one wants to be a fully open participant become impossible to resolve. Not because the CIDS are no in a table somewhere, but because the network gets so congested, they lookups timeout often.

Obviously I've tweaked and tweaked things to make this somewhat work for us. But needless to say, we are not currently embracing a purist IPFS approach, because we simply cant.

We have one repo that is over 40 terabytes alone, and its corrupted too. No way to heal it as well.

I mean... I could go on.