Ask HN: Best pricing model for raw tick by tick cryptocurrency data feed service
I find crypto markets good for small fish like me for many reasons hence this is what I'm focusing on currently.
When looking for good backtesting data sources I found general landscape to be somewhat lacking, there are plenty of daily ohlc datasets, but when it comes to intraday, high resolution raw tick by tick trade and order book data there is not that much choice or expensive and with weird formats or lacking details. And as I'm firm believer of 'garbage in garbage out' rule when it comes to data, I've decided to build my own data collection system. It was supposed to be only for my own internal use and only single exchange, but not sure how and I ended up with supporting most of the top (real volume) exchanges and over 500GB raw compressed data every month. It's all running smoothly after solid couple of months of development and testing, but all that infrastructure is not free, hence I came up with an idea to provide affordable service that would allow others to access that data as well. APIs are really simple (http/REST'ish one that will provide historical exchange streams/datafeed as provided by their's websockets APIs using new lime delimited JSON format) and local 'websocket proxy' that will emulate exchanges websocket APIs - the same subscribe logic as normal exchange API, the same message format - just with nice little feature that is 'time-machine/replay' so one can replay whole exchange's stream from any point in time*.
Anyways, I'm not sure what would be the best billing model for such service and this is question & ask to you all if anyone would be interested in such service would you prefer to have it: on subscription(monthly) basis - price known upfront or maybe one time payment for selected exchange and seleted time periods? or perhaps pay as you go 'metered' billing (eg 0.2$ per million messages received)?
Would you prefer to download such data once and store it all on disk or have constant unlimited API access (no need for large local disks etc, but potentially slower)?
I'd really appreciate your feedback, thanks!