How to setup a query plan in effective at any time for SP or SQL query?

Question

I have a SP which include a group by SQL statement. It retrieve data from a couple of tables which are over 1G size,
When I run this SP at first time, it take more than 5 minutes to get the result. then I run it again and again, Finally, it become very quick, I can get the result within second.
Not sure why. I guess it is because of query plan.
How to make it running at first time to get result within second? How to force a better best query plan in effective at first time to run the query?
If the engine has better plan in memory, could it be lost at some point? because I have the complain from end user said some times it is fast, sometime it is very slow.
How to resolve this problem?

Answer 1

thanks, kevin. Here is the pesudo query( I modify table name as business rule from my company). you are right, mytab3 is a lookup table.
Select d.stock,i.description,c.categoryname,
Round(IsNull(Sum(d.qty),0),2) AS qty,
From mytab1 d,mytab2 s,invent i,mytab3 c       
Where
d.stock != 'param1'
And d.id1 = s.id1    --id1: univarchar(11)        
And i.code = c.code   --code:univarchar(2)         
And d.stock = i.stock  --stock: univarchar(12)           
And i.code2 = d.code2  --code2: univarchar(2)
And d.code2 = 'param2'
And s.id2 = 'param3'   --id2: univarchar(6)
Group By  c.categoryname,d.stock,i.description
Order By d.stock
here is the query plan when run this query:
The command completed with no results returned
QUERY PLAN FOR STATEMENT 1 (at line 1).
Executed in parallel by coordinating process and 4 worker processes.
    STEP 1
        The type of query is SELECT (into Worktable1).
        GROUP BY
        Evaluate Grouped SUM OR AVERAGE AGGREGATE.
        Evaluate Grouped SUM OR AVERAGE AGGREGATE.
        Evaluate Grouped SUM OR AVERAGE AGGREGATE.
        Executed in parallel by coordinating process and 4 worker processes.
        FROM TABLE
            mytab2
            s
        Nested iteration.
        Index : ind_mytab2 _id2
        Forward scan.
        Positioning by key.
        Keys are:
            id2  ASC
        Executed in parallel with a 4-way hash scan.
        Using I/O Size 16 Kbytes for index leaf pages.
        With LRU Buffer Replacement Strategy for index leaf pages.
        Using I/O Size 16 Kbytes for data pages.
        With LRU Buffer Replacement Strategy for data pages.
        FROM TABLE
            mytab1
            d
        Nested iteration.
        Index : ind_det_inv
        Forward scan.
        Positioning by key.
        Keys are:
            id1  ASC
        Using I/O Size 16 Kbytes for index leaf pages.
        With LRU Buffer Replacement Strategy for index leaf pages.
        Using I/O Size 16 Kbytes for data pages.
        With LRU Buffer Replacement Strategy for data pages.
        FROM TABLE
            invent
            i
        Nested iteration.
        Using Clustered Index.
        Index : invent_pk
        Forward scan.
        Positioning by key.
        Keys are:
            stock  ASC
            code2  ASC
        Using I/O Size 2 Kbytes for data pages.
        With LRU Buffer Replacement Strategy for data pages.
        FROM TABLE
            mytab3
            c
        Nested iteration.
        Table Scan.
        Forward scan.
        Positioning at start of table.
        Using I/O Size 2 Kbytes for data pages.
        With LRU Buffer Replacement Strategy for data pages.
        TO TABLE
            Worktable1.
        Parallel work table merge.
    STEP 2
        The type of query is INSERT.
        The update mode is direct.
        Executed by coordinating process.
        Worktable2 created, in allpages locking mode, for ORDER BY.
        FROM TABLE
            Worktable1.
        Nested iteration.
        Table Scan.
        Forward scan.
        Positioning at start of table.
        Using I/O Size 8 Kbytes for data pages.
        With MRU Buffer Replacement Strategy for data pages.
        TO TABLE
            Worktable2.
    STEP 3
        The type of query is SELECT.
        Executed by coordinating process.
        This step involves sorting.
        FROM TABLE
            Worktable2.
        Using GETSORTED
        Table Scan.
        Forward scan.
        Positioning at start of table.
        Using I/O Size 8 Kbytes for data pages.
        With MRU Buffer Replacement Strategy for data pages.
Total estimated I/O cost for statement 1 (at line 1): 1409882.
The sort for Worktable2 is done in Serial