Performance tuning is such a subject that everyone wants to master it. In beginning everybody is at a novice level and spend lots of time learning how to master the art of performance tuning. However, as we progress further the tuning of the system keeps on getting very difficult. I have understood in my early career there should be no need of ego in the technology field. There are always better solutions and better ideas out there and we should not resist them. Instead of resisting the change and new wave I personally adopt it. Let us learn about DB Optimizer.
Here is a similar example, as I personally progress to the master level of performance tuning, I face that it is getting harder to come up with optimal solutions. In such scenarios I rely on various tools to teach me how I can do things better. Once I learn about tools, I am often able to come up with better solutions when I face the similar situation next time.
A few days ago I had received a query where the user wanted to tune it further to get the maximum out of the performance. I have re-written the similar query with the help of AdventureWorks sample database.
SELECT * FROM HumanResources.Employee e INNER JOIN HumanResources.EmployeeDepartmentHistory edh ON e.BusinessEntityID = edh.BusinessEntityID INNER JOIN HumanResources.Shift s ON edh.ShiftID = s.ShiftID;
User had similar query to above query was used in very critical report and wanted to get best out of the query. When I looked at the query – here were my initial thoughts
- Use only column in the select statements as much as you want in the application
- Let us look at the query pattern and data workload and find out the optimal index for it
Before I give further solutions I was told by the user that they need all the columns from all the tables and creating index was not allowed in their system. He can only re-write queries or use hints to further tune this query.
Now I was in the constraint box – I believe * was not a great idea but if they wanted all the columns, I believe we can’t do much besides using *. Additionally, if I cannot create a further index, I must come up with some creative way to write this query. I personally do not like to use hints in my application but there are cases when hints work out magically and gives optimal solutions.
Finally, I decided to use Embarcadero’s DB Optimizer. It is a fantastic tool and very helpful when it is about performance tuning.
I have previously explained how it works over here.
First open DBOptimizer and open Tuning Job from File >> New >> Tuning Job.
Once you open DBOptimizer Tuning Job follow the various steps indicates in the following diagram.
Essentially we will take our original script and will paste that into Step 1: New SQL Text and right after that we will enable Step 2 for Generating Various cases, Step 3 for Detailed Analysis and Step 4 for Executing each generated case. Finally we will click on Analysis in Step 5 which will generate the report detailed analysis in the result pan.
The detailed pan looks like. It generates various cases of T-SQL based on the original query. It applies various hints and available hints to the query and generate various execution plans of the query and displays them in the resultant. You can clearly notice that original query had a cost of 0.0841 and logical reads about 607 pages. Whereas various options which are just following it has different execution cost as well logical read. There are few cases where we have higher logical read and there are few cases where as we have very low logical read.
If we pay attention the very next row to original query have Merge_Join_Query in description and have lowest execution cost value of 0.044 and have lowest Logical Reads of 29. This row contains the query which is the most optimal re-write of the original query. Let us double click over it.
Here is the query:
SELECT * FROM HumanResources.Employee e INNER JOIN HumanResources.EmployeeDepartmentHistory edh ON e.BusinessEntityID = edh.BusinessEntityID INNER JOIN HumanResources.Shift s ON edh.ShiftID = s.ShiftID OPTION (MERGE JOIN)
If you notice above query have additional hint of Merge Join. With the help of this Merge Join query hint this query is now performing much better than before.
The entire process takes less than 60 seconds.
Please note that it the join hint Merge Join was optimal for this query but it is not necessary that the same hint will be helpful in all the queries. Additionally, if the workload or data pattern changes the query hint of merge join may be no more optimal join. In that case, we will have to redo the entire exercise once again. This is the reason I do not like to use hints in my queries and I discourage all of my users to use the same.
However, if you look at this example, this is a great case where hints are optimizing the performance of the query. It is humanly not possible to test out various query hints and index options with the query to figure out which is the most optimal solution. Sometimes, we need to depend on the efficiency tools like DB Optimizer to guide us the way and select the best option from the suggestion provided.
Let me know what you think of this article as well your experience with DB Optimizer. Please leave a comment.
Reference: Pinal Dave (https://blog.sqlauthority.com)