Cassandra schema for time series of mixed data

Question

I'm new to Cassandra and am trying to create a schema for storing and querying a time series of mixed / heterogeneous data. I have questions on how to model the timestamps and the mixed data.

My problem can be abstracted like this:

• I have on the order of 100 million deployed sensor pods.
• Each pod contains a different mix of sensors, where each sensor generates either double or string readings.
• On each day, a pod generates a list of its sensor readings (e.g. sensor1, sensor2, sensor5).

For each sensor pod, I would like to store its sensors' readings for one day. So I am planning to store data like this:

1. podID : Unique ID for the pod
2. dt: represents one entire day, such as July 10, 2018
3. num_data: map that stores numeric sensor readings, e.g. { "sensor1" : -123, "sensor3" : 123 }
4. str_data: map that stores string sensor readings, e.g. { "sensor2" : "foo", "sensor5" : "bar"}

QUESTION 1: What's the best way to store the date? I searched online and found multiple ways.

The first way is use a timestamp as the clustering key:

CREATE TABLE time_series (
    podID     text,
    dt        timestamp,
    num_data  map<text, double>,
    str_data  map<text, double>,
    PRIMARY KEY (podID, dt)
);

The second way I've found is to store the date as individual integers for year, month, and day.

CREATE TABLE time_series (
    podID     text,
    year      int,
    month     int,
    day       int,
    num_data  map<text, double>,
    str_data  map<text, double>,
    PRIMARY KEY (podID, (year, month, day))
);

A third way puts a part of the timestamp into the partition key.

PRIMARY KEY ((podID, month), dt))

How do I choose which way to go?

QUESTION 2: Is my use of map<> data structures the best way to store mixed / varying / heterogeneous types of data?