Here's some additional info that might be helpful. I was following the blog post here on serial vs. parallel optimization thresholds.
Here is info from the final statement in the serial query plan under CL130 (the SELECT):
You can see that the estimated subtree cost is 8.0343, which is above the cost threshold for parallelism (which is set at the default of 5). You can also see that the estimated DOP is 8, but the actual DOP is 1 at execution time.
Then I used the “USE HINT('ENABLE_PARALLEL_PLAN_PREFERENCE')” hint to nudge the CL130 query to use a parallel plan. Here is info from the final statement in the parallel query plan (the SELECT):
Here you can see that the estimated subtree cost is 5.31058, and that the estimated and actual DOP is 8.
So, to summarize:
• The serial plan cost is above our threshold for parallelism (8.0343 > 5)
• The parallel plan cost is below the serial plan cost (5.31058 < 8.0343)
• We know that the query is able to be run in parallel b/c it goes parallel with the hint and runs in parallel under CL120
• But CL130 still chooses the serial plan.