Oddbean new post about | logout
 most of this is ture, except that DHT skepticism is as usual very lazy. Saying that torrent doesn't work without Trackers (that mostly control peers quality and quotas) is wrong (of course they do) but also irrelevant, the DHT's job is to find you peers (or in the case of Pkarr mutable arbitrary data), and Mainline does this job perfectly, as in faster and more reliable than you have any right to expect.

As for "what happens when Mainline hosts billions of records" ... Nothing, each node only hosts the maximum it configured and only responds to as many requests it can handle or just crash and churn away.

The same will apply to Nostr relays if you want them to host the profile metadata of billions of users.

So you tell me, what is more likely to gracefully cope with scale? billions of small nodes that are running without their owners noticing anything. with automatic sharing and redundancy and routing table healing. OR handful of expensive to run servers?

To conclude: In theory, and in practice Mainline is orders of magnitude more capable to do this specific job than Nost relays. If anything needs to face Skepticism it is the newer smaller more expensive to run network of servers.

It is amazing to me, that you think decentralizing search is doable, but a DHT can't scale.

That being said, while there will never be decentralised censorship resistant search, there will always be permissionless search and discovery, because just like the web, anyone can find homeservers and crawl them and run their own Indexer. Don't confuse my pessimism towards decentrlaizing search, with whether or not others will be able to permissionlessly index whatever they want. Of course they can, and nothing in hell we can do to stop them.

Finally, yes homeservers are not too hard, but they are a bit better than Blossom, because they are more like S3 API, with filenames and list pagination which makes them useful for more apps, as a key value database, not just blob storage. 

Other than insisting to pretend that DHTs won't scale (when they already do) I love that you actually understand the layers pretty clearly. 

nostr:note1qqqtsufjm9mfmz4nzjvn2nej78q68jcl0eqmgz3zen8cnw3rez3qekdh9z