Fenwick trees were invented for context modeling with arithmetic coding. Reads and writes are perfectly balanced. For every symbol, there's one write to update the model and one read to determine the cumulative probability.