Seems like an interesting paper, I will give it a read later.
This is true, however I see this as a new window for innovation, not as a call to dismiss what we already have. This does not discard in any way the usefulness of the non-distributed databases and the truth is that a lot of projects still can get away with a simple non-distributed database, as they are inherently much simpler to reason about.
Well sure, I wasn’t advocating for throwing away Postgres tomorrow! You will note I did not cite any distributed databases as alternatives, because there aren’t any! Most of them are closed source, and those that aren’t may well be tomorrow. There is nothing on the market like Postgres (a long history of open development). I use Postgres too!
But the point I was (snarkily) making is that Postgres’s architecture was essentially considered obsolete by its own creator (who is also one of the world’s leading experts) twenty years ago! The “window for innovation” is so far gone you can hardly see it in the rear-view mirror. NewSQL has been a thing for like 15+ years!
I know that everyone is sensitive about Postgres criticism because it evokes the MongoDB web-scale clownery of the 2010s, but just because they were wrong doesn’t mean we can never ever try to do better again.
I agree with you - 99% of webapps could probably run on 1% of a single core, too! But we still write our apps in Elixir just in case we ever want more, right?
It will be the same with databases, but only once someone actually eclipses Postgres in features and stability (of development - no license rugpulls). Which unfortunately nobody has, even after 20 years.
HStore, if you’re wondering, became a closed-source DB (VoltDB), and was therefore relegated to obscurity forever just like every other closed-source DB in history with the sole exception of Oracle.
To steer this discussion back on topic, I want to address this in particular. The funny thing about the HStore paper is that one of the key points made about the “traditional” RDBMS architecture is that it takes an insane amount of complexity to squeeze parallelism out of a fundamentally single-threaded data-structure.
This is NOT simple to reason about. It is, in fact, a total nightmare. The entire tree has to be carefully protected by mutexes (latches in DB parlance) to prevent concurrent access.
The “NewSQL” (HStore) architecture takes the opposite approach: split the database into shards, and then access each shard from a single thread. This is easier to reason about (simpler), and as it turns out it’s also way faster because accessing a fundamentally single-threaded data structure (tree) in parallel is extremely wasteful (my LinkedIn comment…).
What’s funny, though, is if you’ve ever listened to any of Joe Armstrong’s Erlang talks (I’m sure they are beloved here, how could they not be), you’ve heard him make this exact point like a hundred times! I always loved how he put it: (paraphrasing) “shared memory breaks the laws of physics”.
And so, you see, Elixir is actually a pretty interesting language for implementing such a database, as we inherit all of the message passing and parallel scheduling goodies from our Erlang forefathers.
The one thing missing is tools which allow us to build consistent distributed systems. The reason those tools are missing is because OTP is ancient and nobody was doing this stuff back then. Note for example the deafening absence of a Paxos/Raft implementation in the standard library.
And that, really, is what I think is missing from the ecosystem. Consistent distributed systems. In particular, a database!
I love that Elixir allows us to dramatically overreact to any problem lol
Make it a whole ass process, it’s free
Totally agreeing with you though, Elixir in the desktop will be just perfect. The last frontier
Same, at first I thought what is wrong with these people, and then I was like wait, that makes sense 
I don’t like having to deal with setting values within session through a js hook while using liveview:
https://medium.com/elixir-labs/seamless-session-management-in-phoenix-liveview-bridging-the-gap-f0b93c6e7eef
I’m currently looking for an XML solution that supports XPath, is not abandonned, and does not parse to atoms.
We need an Ecto driver supporting Oracle which is till now the number one database for many enterprises including many legacy systems. This will make it easier to introduce other libraries of Elixir like Phoenix, Broadway,.. Etc