Excellent article, i have a query that takes 5 min o more, with this change he takes 4 seconds.

thanks.

Very nice article…clearly differnciated the difference.
But there is one thing that i would like to ask..

As UNION sorts the o/p records on which column it will sort if there are more than one column which do not contain the primary key column??

Pls reply.Thanks in advance.

for e.g., instead of using Difference and vs in the same line

“Difference Between Union vs. Union All – Optimal Performance Comparison”

you could use

“Difference Between Union and Union All – Optimal Performance Comparison”

Its a wonderful article though!

I could also do
SELECT DISTINCT * FROM (
SELECT * FROM A
UNION ALL
SELECT * FROM B
) x

I am trying to tune a query which uses
SELECT DISTINCT * FROM A
UNION
SELECT DISTINCT * FROM B
In this case, the DISTINCTs do not serve any purpose. I am trying to figure out how I should rewrite it.

Any guidelines on which one is better under which circumstances?
Thanks

Keep in mind that these percentages provide no indication of the actual processing time involved, they just indicate the relative amounts of time each operation will take.

But, if we assume that the table scans in UNION, and the table scans in UNION ALL take the same amount of actual time, then we can try to estimate how much more time the UNION operation will take.
Let’s assume the UNION ALL takes 1000msec. Each table scan takes 500msec since each is 50% of the execution.
Now if we go over to UNION and assume that a table scan will take 500msec here too, this 500msec actually only corresponds to 18% of the execution. So, 500msec/.18 = 2778msec.
Thus, the UNION should take approx 2.7 times as long, if the assumption holds true and the execution plan is accurate.

Thanks.

superb article…..

thank u 4 providing these kind of articles…

Nice article, helped a lot in performance tuning the queries. Just had a question, If a query can we written in three ways :
(1) Having an “OR” operator
(2) Having an “Union ALL” operator
(3) having an “IN” operator

Then which method should be used, keeping performance in mind.

Regards,
Deepesh

the percentages are relative , which means that 2×18% of resource in table scan v/s 63% for distinct sort.

you cannot compare %values themselves for 2 different queries.

the point to note is distinct sort is using more resources compared to no distinct sort.

In simple words, read 50% as ’50 % of overall CPU resources/time’ for that execution plan compared to 18%.

For example,

Create table employees ( eid int, ename varchar(10))

insert into employees values ( 1, ‘imran’)
insert into employees values ( 2, ‘Taher’)

We inserted two employees information.

– Lets create another table that stores salary information

create table Salary ( eid int, salary money)

insert into Salary values ( 1, 1000)
insert into Salary values ( 2, 2000)

– We inserted two employees salary information.

–Now if I want to get the complete data out of these two tables, I will use below script

select A.eid
,A.ename
,B.salary
from employees A
join Salary B on A.eid = B.eid

You need to have one common column on which you can join multiple tables.

~ IM.

http://www.pchenry.com/Home/tabid/36/EntryID/166/Default.aspx

I hope you don’t mind but I mentioned this blog, I hope that’s ok with you? Keep up the great work with your blog entries!

I have a question – on query cost diagram it shows 50% for table scan for UNION ALL and 18% for UNION. Why is it so different? Because of the number of records?

If yes then UNION should be faster if it saves so much more resources on that stage?

Thnk you in advance!