Posted by jim_geissman on 09/29/86 11:33

I'm cleaning data which involves updating ~12 million rows with three
different models, progressively. First clean values using the model
with finest granularity, then the remainder with the next model,
finally what's left using the last model. The first model sets ~1/2 of
the rows, the second ~1/4, the third ~1/5, and the remaining 5% don't
get updated.

It's something like this:

UPDATE t SET value=value*m.AdjustmentFactor, updateFlag='updated'
FROM Table t JOIN Models m ON ....
WHERE m.ModelID='first model' AND t.updateFlag IS NULL

Start with 'first model' then 'second model' etc.

I'm wondering what happens if I submit all three queries together, or
as three separate submissions, waiting for the one before to complete.
If I do them all as one group, the query planner might plan for the
second and third updates based on the initial distribution of values.
However, the first update removes half of the rows from consideration,
so it seems to me a new plan should be prepared for the second query,
based on the distribution at that time. If I highlight the queries in
Query Analyzer and execute, are all three plans created at the
beginning? Does putting GO between them (which I do) make any
difference?

This is SQL Server 2000.

Thanks,
Jim

[Back to original message]