SSIS:Best Practices for Better Performance

SQL Server Integration Services (SSIS) 10 Quick Best Practices

Here are the 10 SSIS best practices that would be good to follow during any SSIS package development

§  The most desired feature in SSIS packages development is re-usability. In other ways, we can call them as standard packages that can be re-used during different ETL component development. In SSIS, this can be easily achieved using template features. SSIS template packages are the re-usable packages that one can use in any SSIS project at any number of times. To know more about how to configure this, please see http://support.microsoft.com/kb/908018

§  Avoid using dot (.) naming convention for your package names. Dot (.) naming convention sometime confuses with the SQL Server object naming convention and hence should be avoided. Good approach would be to use underscore (_) instead of using dot. Also make sure that package names should not exceed 100 characters. During package deployment in SQL Server type mode, it is noticed that any character over 100 are automatically removed from package name. This might result your SSIS package failure during runtime, especially when you are using ‘Execute Package Tasks’ in your package.

§  The flow of data from upstream to downstream in a package is a memory intensive task, at most of the steps and component level we have to carefully check and make sure that any unnecessary columns are not passed to downstream. This helps in avoiding extra execution time overhead of package and in turn improves overall performance of package execution.

§  While configuring any OLEDB connection manager as a source, avoid using ‘Table or view’ as data access mode, this is similar to ‘SELECT * FROM <TABLE_NAME>, and as most of us know, SELECT * is our enemy, it takes all the columns in account including those which are not even required. Always try to use ‘SQL command’ data access mode and only include required column names in your SELECT T-SQL statement. In this way you can block passing unnecessary columns to downstream.

§  In your Data Flow Tasks, use Flat File connection manager very carefully, creating Flat File connection manager with default setting will use data type  string [DT_STR] as a default for all the column values. This always might not be a right option because you might have some numeric, integer or Boolean columns in your source, passing them as a string to downstream would take unnecessary memory space and may cause some error at the later stages of package execution.

§  Sorting of data is a time consuming operation, in SSIS you can sort data coming from upstream using ‘Sort’ transformation, however this is a memory intensive task and sometime result in degrade in overall package execution performance. As a best practice, at most of the places where we know that data is coming from SQL Server database tables, it’s better to perform the sorting operation at the database level where sorting can be performed within the query. This is in fact good because SQL Server database sorting is much refined and happens at SQL Server level. This in turn sometime results overall performance improvement in package execution.

§  During SSIS packages development, most of the time one has to share his package with other team members or one has to deploy same package to any other dev, UAT or production systems. One thing that a developer has to make sure is to use correct package protection level. If someone goes with the default package protection level ‘EncryptSenstiveWithUserKey’ then same package might not execute as expected in other environments because package was encrypted with user’s personal key. To make package execution smooth across environment, one has to first understand the package protection level property behaviour, please see http://msdn2.microsoft.com/en-us/library/microsoft.sqlserver.dts.runtime.dtsprotectionlevel.aspx . In general, to avoid most of the package deployment error from one system to another system, set package protection level to ‘DontSaveSenstive’.

§  It’s a best practice to take use of Sequence containers in SSIS packages to group different components at ‘Control Flow’ level. This offers a rich set of facilities

o   Provides a scope for variables that a group of related tasks and containers can use

o   Provides facility to manage properties of multiple tasks by setting property at Sequence container level

o   Provide facility to set transaction isolation level at Sequence container level.

For more information on Sequence containers, please see http://msdn2.microsoft.com/en-us/library/ms139855.aspx.

§  If you are designing an ETL solution for a small, medium or large enterprise business need, it’s always good to have a feature of restarting failed packages from the point of failure. SSIS have an out of the box feature called ‘Checkpoint’ to support restart of failed packages from the point of failure. However, you have to configure the checkpoint feature at the package level.  For more information, please see http://msdn2.microsoft.com/en-us/library/ms140226.aspx.

§  Execute SQL Task is our best friend in SSIS; we can use this to run a single or multiple SQL statement at a time. The beauty of this component is that it can return results in different ways e.g. single row, full result set and XML. You can create different type of connection using this component like OLEDB, ODBC, ADO, ADO.NET and SQL Mobile type etc. I prefer to use this component most of the time with my FOR Each Loop container to define iteration loop on the basis of result returned by Execute SQL Task. For more information, please see http://msdn2.microsoft.com/en-us/library/ms141003.aspx & http://www.sqlis.com/58.aspx.

This article intends to cover the performance improvement techniques and performance constraint scenarios based on the developer’s scope only.

Choice of Transformations

In a real world we would have to do several transformations before the data is actually loaded. The transformations use the buffer memory which in turn affects the performance.So it is very important to understand which transformations influence the performance and how

The transformations can be categorized in to Fully Blocking Transformations, Semi-Blocking Transformations and Non – Blocking transformations.

Fully Blocking Transformations: Blocks the entire dataset to perform the transformation.

Semi-Blocking: Blocks group of data to perform the transformations.

Non – Blocking: No blocking of datasets.

As a general rule, we should try to reduce the number of blocking and semi-blocking transformations.

Non-Blocking transformations	Semi-blocking transformations	Blocking transformations
Audit	Data Mining Query	Aggregate
Character Map	Merge	Fuzzy Grouping
Conditional Split	Merge Join	Fuzzy Lookup
Copy Column	Pivot	Row Sampling
Data Conversion	Unpivot	Sort
Derived Column	Term Lookup	Term Extraction
Lookup	Union All
Multicast
Percent Sampling
Row Count
Script Component
Export Column
Import Column
Slowly Changing Dimension
OLE DB Command

Extra Columns

The most common mistake one does while starting to develop an ssis package is to choose all the columns even if some of them are not required. This might not really sound like a big deal. Consider a scenario where you need to use two fields from a source table which has hundred odd fields. The dataset uses much more buffer size than actually required.

Ensure you always select only those columns which are requires. SSIS by default shows warning messages of column names which are not used.

Configuring Look ups

One of the most common transformations used in SSIS .

The default lookup query for the Lookup Transform is

SELECT * FROM …

The look up allows you to select the table for a look up or a sql query. It is always advisable to use a sql query and only choose the respective columns.

Enabling full caching in look ups enhances the performance of the transformation. However this works only if there are no duplicate records. Another common issue occurs with the blank spaces in the fields for look up. The look up returns no matching data. Its better you trim the fields to get matching records in full cache mode.

Using of SCD

The Slowly changing dimensions are used normally for insert, update or delete records in the table based on the source table data.An alternative approach for this purpose could be done in sql query if both the source and destination is in the same server or through linked servers.

The merge functionality in SQL server 2008 onwards lets you do just that. Also you could write a join query to find those matching records which needs to be changes.

Configuring Source component

Some of the transformations can be totally avoided if they are performed in the source component. Joining two tables from same server, filtering data , sorting data or grouping them can be performed in simple sql query in the source components.

Configuring Destination  component

Fast load vs. normal load

The difference is simple, the former is bulk insert while the later is a row by row insert.(Use a SQL  profiler to see the difference) If you are quite sure about the data that is being processed and if you want to considerably reduce the time taken for huge data insert then Fast load is THE ONE which you need to do. However there are some draw backs, you can’t divert the specific error rows. This is because when there is an error the entire bulk fails. A work around for this is to redirect that failed batch to another destination and do a row by row insert to the same table and get the error record redirected.

OLEDB Adaptor vs. SQL adaptor

If the package is executed on the same machine then using SQL Server adaptor as destination improves the performance considerably.

SSIS Properties

The buffer used for DFT can be altered by the properties DefaultBufferMaxSize and DefaultBufferMaxRows.By increasing them the number of buffers through the data flow. However this should not be increased too much which in turn affects the disk space and does not serve the purpose.

Parallel execution of tasks can be increased by the property MaxConcurrentExecutables.Along with the EngineThreads propery which controls the number of worker threads you need to figure out the right number of parallel executables.

Configuring Flat File Source

While using flat file source it is important we don’t do any unnecessary conversions of the columns .By default all the data are read as strings, so it is important you convert only those columns which require conversions to other type. (Including nvarchar to varchar)

 FastParse indicates whether the column uses the quicker, but locale-insensitive, fast parsing routines that Integration Services provides 

Setting the Fast Parse option in flat file source improves the performance by 7 to 20 % in large files.

Usage of Indexes

I can’t conclude the performance chapter without a mention on the usage of indexes. The Indexes could be a huge constraint while inserting high volumes of data in to tables with several indexes. A work around is to drop and recreate the indexes while inserting data.

Also ensure any unused indexes should be removed from the table.

On the other hand indexes are useful in the source table. Hence put some thought while creating /deleting them.

Performance Improvement in SSIS

(E) EXTRACT IMPROVEMENT

1) If there is a Flat file source / Derived Column Transformation, then set "Fast Parse" to "True.
    - It is available only in Flat File Source & Derived Column Transformation
    - It is specified at column level
    - Default value is False,
    - When we set it true, it will avoid some kind of pre-execute validations and considers all your data fine

    Steps

1. Right-click the Flat File source or Data Conversion transformation, and then click Show Advanced Editor.
2. In the Advanced Editor dialog box, click the Input and Output Properties tab.
3. In the Inputs and Outputs pane, click the column for which you want to enable fast parse.
4. In the Properties window, expand the Custom Properties node, and then set the FastParse property to True.
5. Click OK.

2) Set packet size to 32767 for Connection Manager.
    - This will bump up the packet size from 4K (which is default)
    - This needs network admin to enable "Jumbo Frames"

3) In OLEDB source, use T-SQL Query instead of table as a direct input 
    - This will allow you to choose specific columns instead of pulling all the columns
    - We can specify nolock which avoids locking the table
    - We can use sort, group by, joins, forumlated columns instead of using different transformations like Sort,       
       Merge Join, Derived Column, Aggregate transformations.

4) In Cache connection manager, try to use create a file instead of using memory

5) If same OLEDB source connection, you are using at multiple places, then set "RetainSameConnection" property to "True"

     - This will allow engine to use the same connection every time
     - Default value : False. This will create connection - get data - close connection every time. 
     - by making it to TRUE, above activities will be done only once.

6) Divide source into a chunk instead of having a single master pool.  

(T) TRANSFORM IMPROVEMENT

1) Use Transformation based on the usage and buffer matrix

Behind the scenes, the data flow engine uses a buffer-oriented architecture to efficiently load and manipulate datasets in memory.

• Row Transformations - 
  - They either manipulate data / create new fields using the data that is available in that row. 
  - They might create new columns but not new rows
  - Each output row has a 1:1 relationship with an input row
  - Also known as synchronous transformations
  - Uses existing buffer rather than new buffer
  - Examples -  Derived Column, Data Conversion, Multicast, and Lookup. 
  
  
• Partially blocking transformations 
  - They are often used to combine datasets using multiple data inputs. 
  - As a result, their output may have the same, greater, or fewer records than the total number of input records. 
  - Also known as asynchronous transformations. 
  - Output of the transformation is copied into a new buffer and a new thread may be introduced into the data flow
  - Examples - Merge, Merge Join, and Union All. 
  
  
• Blocking transformations 
  - They must read and process all input records before creating any output records. 
  - They perform the most work and can have the greatest impact on available resources rather than above 2 categories
  - Also known as asynchronous transformations. 
  - With partially blocking transformations, the output of the transformation is copied into a new buffer and a new thread may be introduced into the data flow
  - Example - Aggregate and Sort. 
  
  
  
  

2) Use MERGE statement when you have to do INSERT-UPDATE-DELETE from source to target instead of using multiple transformations

3) Choice of type of cache inside Lookup Transformation
     - Full Cache : for small dataset
     - No Cache : for volatile dataset
     - Partial Cache : for large dataset

4)  Sort, Merge Join, Union All, Pivot, Aggregation SCD, Data Conversion can be easily replaced by normal T-SQL. 

      - There will be much more control on all the objects
      - T-SQL operation will be much more faster than SSIS Transformations because all the 
         buffers won't be used.

5) Make datatype as narrow as possible so that they will allocate less memory

(L) LOAD IMPROVEMENT

1) Try to execute the package on your destination server, rather than source server. 

      - LOAD is expensive operation than EXTRACT
      - So we can execute the package on the same server as destination server

2) Make a smart choice between Dropping/Keeping Index

      - It is not necessary to keep index always OR drop index always before you load.
      - If there is a clustered index, don't drop because data is sorted using this key. And dropping  
        and rebuilding clustered index will take even more time.
      - If there is a single non-clustered index and you expect more than 100% new data, then 
        dropping and re-creating index will help.
      - If there are multiple non-clustered index, probably leave them as it is.

   But these are not thumb rules, trial and error will always give you the best result.

3) If there is a huge huge load on destination, probably partitioning a table will help

4) If there is a huge huge load on destination, probably partitioning a table will help

5) Setting proper value of "Rows per batch" &  "Maximum Insert Commit Size"

Rows per batch - how many rows you want to send to insert the data
Maximum insert Commit Size -  how may rows you want to commit in one shot
      - If the value is 2147483647, these many rows will be committed in one single transaction and 
        they will be committed.
      - If you really have these many rows to load, better you define proper value in this commit 
        size. Let's say if you define 100000, then 1 lac rows will be committed in one shot. A huge 
        DML operation in one single transaction will degrade the performance.
      - If it is 0, it means, a package might stop responding, if the same table is being used by 
        some other source.