Cassandra data model: 1 or 2 column families?

modeling-cows.txt

Entity model:
 - There exist pastures and cows
 - Pastures have many cows
 - Every cow has exactly one pasture
 - Pastures have 2 unique ids a piece - one that rancher A identifies them by, one that rancher B identifies them by

Queries required:
 - given a cow id, look up its type-A pasture id.
 - given a type-A pasture id, look up its type-B pasture id.


Schema possibility 1: 2 column families.
  1. row key = cow id, one static column "a", column value = type-A pasture id
  2. row key = type-A pasture id, one static column "b", column value = type-B pasture id

Schema possibility 2: 1 column family with a secondary index.
  row key = cow id, two static columns "a" and "b", with values that are the pasture id for that type
  make a secondary index on the "a" column, so that you can ask for rows whose type-A pasture id is "x", and get the type-B pasture id directly from the row.

Which is better? Possibility 2 is quite denormalized, since every new cow row for a given pasture will have both types of pasture ids for the pasture. However, the secondary index is easier to maintain than another column family which must be kept in sync.