For your use case of key-value data, where the value is simple, MongoDB is more than you need. It is a document-oriented data store for complex value types. A specific key-value store would suit your needs. RIAK is the one I've looked at though several others exist.
Since your retrieval will be by time range i.e. key range, Elastic's full-text capabilities will, again, be overhead for which you have no payback.
You don't say what your scale out requirement is, whether local write and global read or full replication of the whole data set etc. Several relational engines offer distributed features. MariaDB has Galera, for example. SQL Server has distributed, partitioned views. Other popular systems can afford similar capabilities. NeoDB is relational and multi-host by design. The relational model is based on theoretically sound principles and has decades of industry development. You would be doing yourself a disservice by discounting it out of hand.
And your data does have structure, even if it is a simple structure. You have a primary key (device & time) and a functionally dependent attribte. Whether you define this declaratively in a schema or implicitly in your program's data structures, it still exists.
