Some performance issue about document type

Hi Experts,
Currently, we are thinking moving our search application to ES, but very clear if the feature of ES can meet our requirement?

We have about 1PB data. We used to classify the data by date, newer data can have more hardware resources (e.g more disks )to speed up search, and we delete data by date, we just keep data of the last three years.

So we want to use _type to classify data, the _type is the date( e.g _type=20160416). We can config different number_of_replicas, and we can remove the whole index by removing index under the specific _type.
But if we search across thousands of _types(2013/04/16-2016/04/18, that's thousands of types ) , may it have performance Issues ? And how can we search data from 20150418 to 20160418 by commiting one http request .

Or better way the achieve the goal?

Thanks alot.

I would recommend to use different indices per timeframe (with only one type each) rather than different types.