How can I speed up this linq to sql function?

Question

I have a function (called "powersearch", the irony!) that searches for a set of strings across a bunch(~ 5) of fields. The words come in as one string and are separated by spaces.
Some fields can have exact matches, others should have "contains".

(Snipped for brevety)

//Start with all colors
IQueryable<Color> q = db.Colors;
//Filter by powersearch
if (!string.IsNullOrEmpty(searchBag.PowerSearchKeys)){
    foreach (string key in searchBag.SplitSearchKeys(searchBag.PowerSearchKeys)
                                    .Where(k=> !string.IsNullOrEmpty(k))){
        //Make a local copy of the var, otherwise it gets overwritten
        string myKey = key;
        int year;
        if (int.TryParse(myKey, out year) && year > 999){
            q = q.Where(c => c.Company.Name.Contains(myKey)
                || c.StockCode.Contains(myKey)                                
                || c.PaintCodes.Any(p => p.Code.Equals(myKey))
                || c.Names.Any(n => n.Label.Contains(myKey))
                || c.Company.CompanyModels.Any(m => m.Model.Name.Contains(myKey))
                || c.UseYears.Any(y => y.Year.Equals(year))
            );
        }
        else{
            q = q.Where(c => c.Company.Name.Contains(myKey)
                || c.StockCode.Contains(myKey)
                || c.PaintCodes.Any(p => p.Code.Contains(myKey))
                || c.Names.Any(n => n.Label.Contains(myKey))                                
                || c.Company.CompanyModels.Any(m => m.Model.Name.Equals(myKey))
            );
        }
    }
}

Because the useYear count is rather large, I tried to check for it as little as possible by outruling all numbers that can never be a number that makes sence in this case. Similar checks are not possible on the other fields since they can pretty much contain any thinkable string.

Currently this query takes about 15 secs for a single, non-year string. That's too much. Anything I can do to improve this?

--Edit--
Profiler shows me the following info for the part where the string is not a year:

exec sp_reset_connection Audit login

exec sp_executesql N'
SELECT COUNT(*) AS [value]
FROM [dbo].[CLR] AS [t0]
INNER JOIN [dbo].[CO] AS [t1] ON [t1].[CO_ID] = [t0].[CO_ID]
WHERE 
 ([t1].[LONG_NM] LIKE @p0)
 OR ([t0].[EUR_STK_CD] LIKE @p1)
 OR (EXISTS(
  SELECT NULL AS [EMPTY]
  FROM [dbo].[PAINT_CD] AS [t2]
  WHERE ([t2].[PAINT_CD] LIKE @p2)
   AND ([t2].[CLR_ID] = [t0].[CLR_ID])
   AND ([t2].[CUSTOM_ID] = [t0].[CUSTOM_ID])
  )
 )OR (EXISTS(
  SELECT NULL AS [EMPTY]
  FROM [dbo].[CLR_NM] AS [t3]
  WHERE ([t3].[CLR_NM] LIKE @p3)
   AND ([t3].[CLR_ID] = [t0].[CLR_ID])
   AND ([t3].[CUSTOM_ID] = [t0].[CUSTOM_ID])
  )
 ) OR (EXISTS(
  SELECT NULL AS [EMPTY]
  FROM [dbo].[CO_MODL] AS [t4]
  INNER JOIN [dbo].[MODL] AS [t5] ON [t5].[MODL_ID] = [t4].[MODL_ID]
  WHERE ([t5].[MODL_NM] = @p4)
   AND ([t4].[CO_ID] = [t1].[CO_ID])
  )
 )
',N'@p0 varchar(10),@p1 varchar(10),@p2 varchar(10),@p3 varchar(10),@p4 varchar(8)',@p0='%mercedes%',@p1='%mercedes%',@p2='%mercedes%',@p3='%mercedes%',@p4='mercedes'

(took 3626 msecs) Audit Logout (3673 msecs) exec sp_reset_connection (0msecs) Audit login

exec sp_executesql N'
SELECT TOP (30) 
[t0].[CLR_ID] AS [Id],
[t0].[CUSTOM_ID] AS [CustomId],
[t0].[CO_ID] AS [CompanyId], 
[t0].[EUR_STK_CD] AS [StockCode], 
[t0].[SPCL_USE_CD] AS [UseCode], 
[t0].[EFF_IND] AS [EffectIndicator]
FROM [dbo].[CLR] AS [t0]
INNER JOIN [dbo].[CO] AS [t1] ON [t1].[CO_ID] = [t0].[CO_ID]
WHERE 
 ([t1].[LONG_NM] LIKE @p0)
 OR ([t0].[EUR_STK_CD] LIKE @p1)
 OR (EXISTS(
  SELECT NULL AS [EMPTY]
  FROM [dbo].[PAINT_CD] AS [t2]
  WHERE ([t2].[PAINT_CD] LIKE @p2)
   AND ([t2].[CLR_ID] = [t0].[CLR_ID])
   AND ([t2].[CUSTOM_ID] = [t0].[CUSTOM_ID])
  )
 )
 OR (EXISTS(
  SELECT NULL AS [EMPTY]
  FROM [dbo].[CLR_NM] AS [t3]
  WHERE ([t3].[CLR_NM] LIKE @p3)
   AND ([t3].[CLR_ID] = [t0].[CLR_ID])
   AND ([t3].[CUSTOM_ID] = [t0].[CUSTOM_ID])
  )
 )
 OR (EXISTS(
  SELECT NULL AS [EMPTY]
  FROM [dbo].[CO_MODL] AS [t4]
  INNER JOIN [dbo].[MODL] AS [t5] ON [t5].[MODL_ID] = [t4].[MODL_ID]
  WHERE ([t5].[MODL_NM] = @p4)
   AND ([t4].[CO_ID] = [t1].[CO_ID])
  )
 )'
,N'@p0 varchar(10),@p1 varchar(10),@p2 varchar(10),@p3 varchar(10),@p4 varchar(8)',@p0='%mercedes%',@p1='%mercedes%',@p2='%mercedes%',@p3='%mercedes%',@p4='mercedes'

(took 3368 msecs)

The database structure, sadly, is not under my control. It comes from the US and has to stay in the exact same format for compatibility reasons. Although most of the important fields are indeed indexed, they are indexed in (unnecessary) clustered primary keys. There's verry little I can do about that.

Answer 1

Okay, let's break this down - the test case you're interested in first is a single non-year, so all we've got is this:

q = q.Where(c => c.Company.Name.Contains(myKey)
            || c.StockCode.Contains(myKey)
            || c.PaintCodes.Any(p => p.Code.Contains(myKey))
            || c.Names.Any(n => n.Label.Contains(myKey))
            || c.Company.CompanyModels.Any(m => m.Model.Name.Equals(myKey))

Am I right? If so, what does the SQL look like? How long does it take just to execute the SQL statement in SQL Profiler? What does the profiler say the execution plan looks like? Have you got indexes on all of the appropriate columns?

Answer 2

Use compiled queries.

If you don't, you will lose up to 5-10x times performance, as LINQ-to-SQL will have to generate SQL from query every time you call it.

Things become worse when you use non-constants in LINQ-to-SQL as getting their values is really slow.

This assumes that you already have indexes and sane DB schema.

BTW, I am not kidding about 5-10x part.