select sum(col_1) from t1;

Let us recap so far:

1. create an index on the column so that you have a very small "duplicate" object to be scanned instead of the actual full table.

2. put it on EXADATA and let SMARTSCAN do its magic of eliminating physical IO, doing parallel physical IO, and reducing network traffic.

3. pre-load it in memory and skip all the physical IO.

4. use compression to reduce the size of the table and hope the savings in IO outweighs the increased CPU needed for decompression (usually works).

------ and I add --------

5. use partitioning and parallel query to divide the task up into many pieces that multiple processes can do at the same time.

6. use a materialized view or other pre-aggregation strategies to pre-sum results for strategically reusable subsets of the rows.

7. redesign your solution so that you don't need a 10 billion row table or (I guess here) only one wide table with 100s of unrelated metrics on a single row.

8. explore your problem space to see if the SUM is really your problem or if it is something else (chained rows, joins to other tables, bug needing a patch (who knows?)).