Would something like spark/storm/etc be simpler to implement on top of Elixir? Possibly more performant? Or make better use of CPU cores?
My own impression (as the author of http://www.kiba-etl.org which is a Ruby ETL framework) is that Elixir is going to be a great fit for high-quality, decent throughput data processing of all kinds (streaming, batching, or other such as websocket-api-endpoint types).
The immutable data + lightweight processes + ability to go distributed + concise code (testable, composable etc) is truly appealing in my opinion.
I will share more stuff online (at http://thibautbarrere.com) when I can, but I can only say this is so far very promising.
>In the streaming case, documents arrive at a very fast rate (e.g. average of 6000 per second in the case of Twitter) and with this kind of velocity and volume it is impractical to build the inverted document index in real-time.
Actually, it's completely practical. It's expensive, but it can be done. When I was at Loggly we built a big ES cluster, pumped 10,000s log messages into it per second, and served queries. Ingest to query was on the order of seconds - it still runs today. The key, of course, is not to build a single inverted index.
I don't want to underestimate the work, but it is practical. And expensive.
http://www.slideshare.net/AmazonWebServices/infrastructure-a...
