Replace Subselect for something more efficient - tsql

Question

I have this query which takes a long time, partly because the number of records in the table excedd 500 000 records, but the join I have to use slows it down quite a lot, at least to my beliefs
SELECT TOP (10) PERCENT H1.DateCompteur, CASE WHEN (h1.cSortie - h2.cSortie > 0)
THEN h1.cSortie - h2.cSortie ELSE 0 END AS Compte, H1.IdMachine
FROM dbo.T_HistoriqueCompteur AS H1 INNER JOIN
dbo.T_HistoriqueCompteur AS H2 ON H1.IdMachine = H2.IdMachine AND H2.DateCompteur =
(SELECT MAX(DateCompteur) AS Expr1
FROM dbo.T_HistoriqueCompteur AS HS
WHERE (DateCompteur < H1.DateCompteur) AND (H1.IdMachine = IdMachine))
ORDER BY H1.DateCompteur DESC
The order by is important since I need only the most recent informations. I tried using the ID field in my sub select since they are ordred by date but could not detect any significant improvement.
SELECT TOP (10) PERCENT H1.DateCompteur, CASE WHEN (h1.cSortie - h2.cSortie > 0)
THEN h1.cSortie - h2.cSortie ELSE 0 END AS Compte, H1.IdMachine
FROM dbo.T_HistoriqueCompteur AS H1 INNER JOIN
dbo.T_HistoriqueCompteur AS H2 ON H1.IdMachine = H2.IdMachine AND H2.ID =
(SELECT MAX(ID) AS Expr1
FROM dbo.T_HistoriqueCompteur AS HS
WHERE (ID < H1.ID) AND (H1.IdMachine = IdMachine))
ORDER BY H1.DateCompteur DESC
the table I use look a little like this (I got much more columns but they are unused in this query).
ID bigint
IdMachine bigint
cSortie bigint
DateCompteur datetime
I think that if I could get rid of the sub select, my query would run much faster but I can't really find a way to do so. What I really want to do is to find the previous row with the same IdMachine so that I can calculate the difference between the two cSortie values. The case in the query is because something it's reseted to 0 and in this case, I want to return 0 instead of a negative value.
So my question is this : Can I do better than what I already have ??? I plan to put this in a view if that makes a difference.

Answer 1

Try this query
WITH T as
(
SELECT TOP (10) PERCENT H1.DateCompteur, H1.cSortie as cSortie1, H1.IdMachine,
(
SELECT TOP 1 H2.cSortie
FROM dbo.T_HistoriqueCompteur H2
WHERE (H2.DateCompteur < H1.DateCompteur) AND (H1.IdMachine = H2.IdMachine)
ORDER BY H2.DateCompteur DESC
) as cSortie2
FROM dbo.T_HistoriqueCompteur AS H1
ORDER BY H1.DateCompteur DESC
)
select DateCompteur,
CASE WHEN (cSortie1 - cSortie2 > 0)
THEN cSortie1 - cSortie2
ELSE 0 END
AS Compte,
IdMachine
FROM T

Answer 2

You could also try CTE's (common table expressions) with windowing functions (ROW_NUMBER):
;WITH CTE AS
(
SELECT ID,IdMachine,cSortie,ROW_NUMBER() OVER(PARTITION BY h.IdMachine ORDER BY ID ASC) AS [ROW]
FROM T_HistoriqueCompteur h
)
SELECT
TOP (10) PERCENT
H1.DateCompteur,
CASE WHEN (h1.cSortie - h2.cSortie > 0) THEN h1.cSortie - h2.cSortie
ELSE 0
END AS Compte,
H1.IdMachine
FROM dbo.T_HistoriqueCompteur AS H1
INNER JOIN CTE cte on cte.idmachine = h1.idmachine and cte.id = h1.id
INNER JOIN CTE h2 on h2.idmachine = cte.idmachine and h2.row + 1 = cte.row
ORDER BY H1.DateCompteur DESC