SQL Server : optimize the efficiency with many joins relationship - sql-server

Question

I have SQL Server code which takes long time to run the result. In the past, it took 15 minutes. But recently, might as a result of accumulated sales data, it took 2 hours to get the result!!
Therefore, I would like to get some advice regarding how to optimize the code:
The code structure is simple: just to get the sales sum for different regions for different time periods and for each SKU. (I have deleted some code here is to find the different SKU for each materials without size).
Many thanks in advance for your help.
The main code structure is as below, since it is almost the same, so I just give the first 2 paragraphs as example:
SELECT SKU from [MATINFO]
-- Global Sales History Qty - All the years
LEFT JOIN
(
SELECT SKU,SUM([SALES Qty]) as [Global Sales History Qty - All the years]
from dbo.[SALES]
where [PO] IS NOT NULL
group by SKU
)histORy
on MATINFO.[SKU]=histORy.[SKU]
-- Global Sales History Qty - Past 2 years
LEFT JOIN
(
SELECT (
SELECT SKU,SUM([SALES Qty]) as [Global Sales History Qty - All the years]
from dbo.[SALES]
where [PO] IS NOT NULL
group by SKU
/* date range */
and ([ORDER DATE] = '2015.11' OR [ORDER DATE] = '2015.12' or [ORDER DATE] like '%2015%' OR [ORDER DATE] like '%2016%' )
group by SKU
)histORy2
on MATINFO.[SKU]=histORy2.[SKU]
--Global Sales History Qty - Past 1 years
......SIMILAR TO THE CODE STRUCTURE AS ABOVE

Answer 1

The most likely cause of the poor performance is using string for dates and possibly the lack if as adequate indexes.
like '%2015%'
Using double-ended wildcards with like results in full table scans so subqueries are scanning the whole table each time you serach for a different date range. Using temp tables will not solve the underlying issues.
[added later]
Another facet of your original query structure might reduce the number of scans you need of the data - by using "conditional aggregates"
e.g. here is a condensed version of your original query
SELECT
SKU
FROM [MATINFO]
-- Global Sales History Qty - All the years
LEFT JOIN (SELECT
SKU
, SUM([SALES Qty]) AS [Global Sales History Qty - All the years]
FROM dbo.[SALES]
WHERE [PO] IS NOT NULL
GROUP BY
SKU) histORy ON MATINFO.[SKU] = histORy.[SKU]
-- Global Sales History Qty - Past 2 years
LEFT JOIN (SELECT
SKU
, SUM([SALES Qty]) AS [Global Sales History Qty - Past 2 years]
FROM dbo.[SALES]
WHERE [PO] IS NOT NULL
/* date range */
AND [ORDER DATE] >= '20151101' AND [ORDER DATE] < '20161101'
GROUP BY
SKU) histORy2 ON MATINFO.[SKU] = histORy2.[SKU]
That requires a 2 complete passes of the data in dbo.[SALES], but if you were to use a case expression inside the SUM() function you need only one pass of the data (in this example)
SELECT
SKU
, SUM([SALES Qty]) AS [Qty_all_years]
, SUM(CASE
WHEN [ORDER DATE] >= '20151101' AND [ORDER DATE] < '20161101'
THEN [SALES Qty]
END) AS [Qty_past_2_years]
FROM dbo.[SALES]
WHERE [PO] IS NOT NULL
GROUP BY
SKU
I suspect you could apply this logic to most of the columns and substantially improve efficiency of the query when coupled with date columns and appropriate indexing.

Answer 2

Expansion on my comment. Note it is just a suggestion, no guarantee if will run faster.
Take the following derived table histORy:
SELECT SKU,SUM([SALES Qty]) AS [Global Sales History Qty - All the years]
FROM dbo.[SALES]
WHERE [PO] IS NOT NULL
GROUP BY SKU
Before you run your query, materialize the derived table in a temporary table:
SELECT SKU,SUM([SALES Qty]) AS [Global Sales History Qty - All the years]
INTO #histORy
FROM dbo.[SALES]
WHERE [PO] IS NOT NULL
GROUP BY SKU
Then use the temporary table in the query:
LEFT JOIN #histORy AS h ON MATINFO.[SKU]=h.[SKU]
In this case you may want to have a index on the SKU field, so you could create the temporary table yourself, slap an index on it, populate with INSERT INTO #history... SELECT ... etc.