SSIS Data Flow Design and Tuning

Now that you know how the Data Flow engine works, it will be easier to understand Data Flow design principles and tuning practices. 

Designing a data-processing solution requires more than just sending the source data into a black-box transformation engine with outputs that push the data into the destination. Of course, system requirements will dictate the final design of the process, including but not limited to the following:

  • Source and destination system impact
  • Processing time windows and performance
  • Destination system state consistency
  • Hard and soft exception handling and restartability needs
  • Environment architecture model, distributed hardware, or scaled-up servers
  • Solution architecture requirements, such as flexibility of change or OEM targeted solutions
  • Modular and configurable solution needs
  • Manageability and administration requirements

Looking at this list, you can quickly map several of these items to what you have learned about SSIS already. In most cases, a good architecture leverages the built-in functionality of the tool, which reduces administrative and support requirements. The tool selection process, if it is not completed before a solution is developed, should include a consideration of the system requirements and the functionality of available products.

Data Flow Design Practices

Keep the following four design practices in mind when creating new packages:

  • Limit synchronous processes
  • Monitor the memory use of blocking and semi-blocking transformations
  • Reduce staging and disk I/O
  • Reduce reliance on an RDBMS

To limit synchronous processes, you should be conscious of processes that need to complete before the next process begins. For example, if you run a long INSERT T-SQL statement that takes one-half hour to complete, and then run an UPDATE statement that updates the same table, the UPDATE statement cannot run until the INSERT script finishes. These processes are synchronous. In this case, it would be better to design a Data Flow that handles the same logic as the INSERT statement and combines the UPDATE logic at the same time (also using a SQL UPDATE); that way, you are not only taking advantage of the Data Flow but making the logic asynchronous. You can seriously reduce overall process times by taking this approach.

As mentioned earlier, blocking and semi-blocking transformations require buffers to be pooled in memory until either the last row is available or a match is received (in the case of a Merge Join). These transformations can be very useful, but you need to ensure that you have sufficient memory on your server in order for them to perform well. If you have a large aggregate operation across 100 million rows, you are likely better off handling this through the SQL Server relational engine.

Reducing disk I/O is about minimizing the staging requirements in your ETL process. Disk I/O is often the biggest bottleneck in an ETL job because the bulk operations involve moving a lot of data, and when you add staging data to a database, that data ultimately needs to be saved to the disk drives. Instead, reduce your need on staging tables and leverage the Data Flow for those same operations; that way, you decrease the disk overhead of the process and achieve better scalability. The Data Flow primarily uses memory, and memory is a lot faster to access than disks, so you will gain significant improvements in terms of speed. In addition, when you stage data to a table, you are doubling the disk I/O of the data, because you are both inserting and retrieving table data.

Keep in mind that solution requirements often drive design decisions, and there are situations where staging or the RDBMS are useful in data processing. Some of these are discussed in this section. Your goal, though, is to rethink your design paradigms with SSIS.

Reducing the RDBMS reliance is similar to reducing staging environments, but it also means reducing the logic you place on the DBMS to perform operations like grouping and data cleansing. This not only reduces the impact on your RDBMS, but when using production databases, it alleviates the load and makes room for more critical RDBMS operations. 

These four principles are worked out further in the next section, which discusses ways to leverage the Data Flow for your ETL operations.

Leveraging the Data Flow

The biggest value that SSIS brings is the power of the Data Flow. Not to minimize the out-of-the-box functionality of restartability, configurations, logging, event handlers, or other Control Flow Tasks, the primary goal of the engine is to “integrate,” and the Data Flow is the key to realizing that goal. Accomplishing data-processing logic through Data Flow Transformations provides enhanced performance and greater flexibility.

Many data architects come from DBA backgrounds, so when they are trying to solve a data integration, processing, or cleansing scenario, their first impulse is to use an RDBMS, such as SQL Server, for ETL operations.

In some ways, moving to SSIS requires thinking in different terms — Data Flow terms. In Reliability and Scalability Topic, you looked at the different Data Flow Transformations, so the focus in this section is on applying some of those components to design decisions and translating the SQL-based designs into Data Flow processes.

In summary, the four architecture best practices described in the preceding section relate directly to the value that the Data Flow provides:

  • Limit synchronous processes: By bringing more of the processing logic into the Data Flow, the natural result is fewer process-oriented steps that require completion before proceeding. In the Reliability and Scalability Topic, you looked at the general streaming nature of the Data Flow. This translates to reduced overall processing times.
  • Monitor the memory use of blocking and semi-blocking transformations: When memory becomes scarce on your server, SSIS begins to copy buffers to disk or spool them to disk. Once this happens, your package slows down dramatically. The most intensive memory transformations are blocking and semi-blocking transformations. However, other transformations, like the cached Lookup, also can require a lot of memory if the Lookup table contains millions of items. All these transformations perform very well until they near the threshold of memory on your server. It is best to monitor the memory to ensure that you avoid a low-memory situation. In particular, throttle the memory of SQL Server if it is on the same server where your SSIS packages are running.
  • Reduce staging and expensive IO operations: The Data Flow performs most operations in memory (with occasional use of temp folders and some interaction with external systems). Whenever processing happens on data that resides in RAM, processing is more efficient. Disk I/O operations rely on the performance of the drives, the throughput of the I/O channels, and the overhead of the operating system to write and read information to and from the disk. With high volumes or bursting scenarios typical with data processing and ETL, disk I/O is often a bottleneck.
  • Reduce reliance on RDBMS: Relational engines are powerful tools, and the point here is not to detract from their appropriate uses to store and manage data. By using the Data Flow to cleanse and join data rather than the RDBMS, the result is reduced impact on the relational system, which frees it up for other functions that may be higher priority. Reading data from a database is generally less expensive than performing complex joins or complicated queries. In addition, related to the first bullet, all RDBMS operations are synchronous. Set-based operations, while they are very useful and optimized in a relational database system, still require that the operation be complete before the data is available for other purposes. The Data Flow, conversely, can process joins and Lookups and other cleansing steps in parallel while the data is flowing through the pipeline. However, note that an RDBMS engine can be leveraged in certain ways; for example, if a table has the right indexes, you can use an ORDER BY, which may be faster than an SSIS Sort Transformation.

Data Integration and Correlation

As discussed in Joining Data Topic, the Data Flow provides the means to combine data from different source objects completely independent of the connection source where the data originates. The most obvious benefit of this is the ability to perform in-memory correlation operations against heterogeneous data without having to stage the data. Said in another way, with SSIS, you can extract data from a flat file and join it to data from a database table inside the Data Flow, without first having to stage the flat file to a table and then perform a SQL Join operation. This can be valuable even when the data is coming from the same source, such as a relational database engine; source data extractions are more efficient without complex or expensive joins, and data can usually begin to flow into the Data Flow immediately.

In addition, single-table SELECT statements provide less impact on the source systems than do pulls where join logic is applied. Certainly there are situations where joining data in the source system may be useful and efficient; in many cases, however, focusing on data integration within the Data Flow will yield better performance. When different source systems are involved, the need to stage the data is reduced.

Several of the built-in transformations can perform data correlation similar to how a database would handle joins and other more complex data relationship logic. The following transformations provide data association for more than one Data Source:

  • Lookup
  • Merge Join
  • Merge
  • Union All
  • Fuzzy Lookup
  • Term Lookup
  • Term Extract

Joining Data Topic describes how to leverage the joining capabilities of SSIS, a great reference for designing your SSIS Data Flows.

Furthermore, beyond the built-in capabilities of SSIS, custom components allow more complex or unique scenarios to be handled. This is discussed in Programming and Extending SSIS Topic.

Data Cleansing and Transformation

The second major area to which you can apply the Data Flow is data cleansing. As discussed in Advanced Data Cleansing in SSIS Topic, cleansing data involves managing missing values; correcting out-of-date, incomplete, or miskeyed data; converting values to standard data types; changing data grain or filtering data subsets; and de-duplicating redundant data. Consistency is the goal of data cleansing, whether the Data Source is a single system or multiple disparate sources.

Many of the Data Flow Components provide data-cleansing capabilities or can participate in a data-cleansing process. Some of the more explicit transformations usable for this process include the following:

  • Aggregate
  • Character Map
  • Conditional Split
  • Data Conversion
  • Derived Column
  • DQS Cleansing
  • Fuzzy Grouping
  • Fuzzy Lookup
  • Pivot
  • Script Component
  • Sort (with de-duplicating capabilities)
  • Unpivot

Each of these transformations, or a combination of them, can handle many data-cleansing scenarios. A few of the transformations provide compelling data-cleansing features that even go beyond the capabilities of many relational engines by using the Data Flow. For example, the Fuzzy Lookup and Fuzzy Grouping (de-duplication) provide cleansing of dirty data by comparing data similarity within certain defined ranges. Pivot and Unpivot have the ability to transform data coming in by pivoting rows to columns or vice versa.

Also, the Script Transformation offers very powerful data-cleansing capabilities with the full features of VB.NET embedded; it is highlighted in detail in Scripting in SSIS Topic. Because the goal of this Understanding and Tuning the Data Flow Engine Topic is to discuss the application of SSIS, the example focuses on a couple of common examples of data cleansing using the Derived Column Transformation and the Aggregate Transformation. These two transformations are particularly relevant to how data cleansing can be accomplished in the Data Flow in comparison with common query logic.

Staging Environments

So far in this Understanding and Tuning the Data Flow Engine Topic, the emphasis has been on thinking in Data Flow terms by moving core data process logic into the Data Flow. In most cases, this will yield high-performance results, especially when the timeliness of moving the data from point A to point B is the highest priority, such as in near real-time or tight-processing-window scenarios. Doing this also mitigates some management overhead, limiting interim database usage.

A few situations merit staging environments and are worth mentioning for consideration:

  • Restartability: The built-in checkpoint logic of SSIS revolves around the Control Flow. This means that a failure in the Data Flow will not persist the data state. Rather, when the package is restarted, the Data Flow will restart from the beginning. The implications affect design if the source system is in flux and an error in the Data Flow causes a processing window to be missed. By landing the raw data first, the chance for data errors is minimized, and in the event of a failure during the load process, the package can be restarted from the staged data.
  • Processing windows and precedence: Certain requirements may dictate that the various source extraction windows do not line up with each other or with the data load window for the destination. In these scenarios, it would be necessary to stage the data for a period of time until the full data set is available or the destination database load window has been reached.
  • Source backpressure: At times, the Data Flow Transformations may apply backpressure on the source extractions. This would happen when the flow of incoming data is faster than the ability of the transformations to handle the data processing in the pipeline. The backpressure created slows down the extraction on the source system, and if the requirement is to extract the data in the fastest time with the least impact, then staging the raw data extract may help eliminate the backpressure.
  • Data Flow optimization: Staging certain elements, such as business keys, can actually provide valuable data to optimize the primary Data Flow. For example, if the Lookup Source query can be filtered based on a set of keys that was prestaged, this may allow overall gains in processing times by reducing the time it takes to load the Lookup plus the amount of memory needed for the operation. A second example is the use of staging to perform set-based table updates. Updates in a large system are often the source of system bottlenecks, and because SSIS cannot perform set-based updates in the Data Flow, one solution is to stage tables that can be used in a later Execute SQL Task for a set-based update, which may provide a more efficient process.

Staged data can also prove useful in data validation and error handling. Given some of the uses of staging, is there a way to accomplish data staging but still retain the performance gain by leveraging the Data Flow? Yes. One suggestion is the reduction of synchronous processing in the Control Flow. When you want to introduce a data-staging environment, it’s natural to first pick up the data from the source and land it to a staging environment, and then pick the data back up from the staging environment and apply the transformation logic to it; but what about landing the raw data to a staging environment at the same time that the transformations are applied? Below screen shot shows a Data Flow designed with a staging table that does not require the data to reside in the table before the transformation logic is applied.

Data Flow designed with a staging table that does not require the data to reside in the table before the transformation logic is applied

The Multicast Transformation in this example is taking the raw source data and allowing it to stream down to the core Data Flow, while at the same time the raw source data is being staged to a table. The data within the table is now available to query for data validation and checking purposes; in addition, it provides a snapshot of the source system that can then be used for reprocessing when needed. Although the data is landed to staging, two differences distinguish this example from a model that first stages data and then uses the staged data as a source.

First, as already mentioned, the process is no longer synchronous; in many cases, data can move from point A to point B in the time it takes to simply extract the data from A. Second, the staging process requires only a single pass on the staging table (for the writes), rather than the I/O overhead of a second pass that reads the data from the staging. If your restartability requirements and source systems allow, this approach may provide the best of both worlds — leveraging the Data Flow but providing the value of a staging environment.

Optimizing Package Processing

There are a few techniques you can apply when you’re streamlining packages for performance. This section covers how to apply certain optimization techniques to achieve better throughput.

Optimizing Buffers, Execution Trees, and Engine Threads

Recall from earlier in this Understanding and Tuning the Data Flow Engine Topic that for each execution tree in a Data Flow, a different buffer profile is used. This means that downstream execution trees may require different columns based on what is added or subtracted in the Data Flow. You also saw that the performance of a buffer within a Data Flow is directly related to the row width of the buffer. Narrow buffers can hold more rows, enabling higher throughput.

Some columns that are used in an execution tree may not be needed downstream. For example, if an input column to a Lookup Transformation is used as the key match to the reference table, this column may not be needed after the Lookup and therefore should be removed before the next execution tree. SSIS does a good job of providing warnings when columns exist in an execution tree but are not used in any downstream transformation or destination component. below screen shot shows the Execution Results tab (also called the Progress tab when in runtime mode) within a package for which column usage has not been optimized in the Data Flow. Each warning, highlighted with a yellow exclamation point, indicates the existence of a column not used later in downstream components and which therefore should be removed from the pipeline after initial use.

the Execution Results tab (also called the Progress tab when in runtime mode)

The warning text describes the optimization technique well:

[SSIS.Pipeline] Warning: The output column “Average_Sale” (7) on
output “Derived
Column Output” (6) and component “Average Calc” (2) is not
subsequently used in
the Data Flow task. Removing this unused output column can increase
Data Flow task

Any component with asynchronous outputs whose input closes out an execution tree has the option of removing columns in the output. You would normally do this through the edit dialog of the transformation, but you can also do it in the Advanced Editor if the component provides an advanced properties window. For example, in the Union All Transformation, you can highlight a row in the editor and delete it with the Delete keyboard key. This ensures that the column is not used in the next execution tree.

A second optimization technique involves increasing processor utilization by adding the more execution threads for the Data Flow. Increasing the EngineThreads Data Flow property to a value greater than the number of execution trees plus the number of Source Components ensures that SSIS has enough threads to use.

Careful Use of Row-Based Transformations

Row-based transformations, as described earlier in this Understanding and Tuning the Data Flow Engine Topic, are nonblocking transformations, but they exhibit the functionality of interacting with an outside system (for example, a database or file system) on a row-by- row basis. Compared with other non-blocking transformations, these transformations are slower because of this behavior. The other type of nonblocking transformation, streaming, can use internal cache or provide calculations using other columns or variables readily available to the Data Flow, making them perform very fast. Given the nature of row-based transformations, their usage should be cautious and calculated.

Of course, some row-based transformations have critical functionality, so this caution needs to be balanced with data-processing requirements. For example, the Export and Import Column Transformation can read and write from files to columns, which is a very valuable tool, but it has the obvious overhead of the I/O activity with the file system.

Another useful row-based transformation is the OLE DB Command, which can use input column values and execute parameterized queries against a database, row by row. The interaction with the database, although it can be optimized, still requires overhead to process. below screen shot shows a SQL Server Trace run against a database that is receiving updates from an OLE DB Command.

SQL Server Trace run against a database that is receiving updates from an OLE DB Command.

This is only a simple example, but for each row that goes through the OLE DB Command, a separate UPDATE statement is issued against the database. Taking into consideration the duration, reads, and writes, the aggregated impact of thousands of rows will cause Data Flow latency at the transformation.

For this scenario, one alternative is to leverage set-based processes within databases. In order to do this, the data needs to be staged during the Data Flow, and you need to add a secondary Execute SQL Task to the Control Flow that runs the set-based update statement. The result may actually reduce the overall processing time when compared with the original OLE DB Command approach. This alternative approach is not meant to diminish the usefulness of the OLE DB Command but rather meant to provide an example of optimizing the Data Flow for higher-volume scenarios that may require optimization.

Understanding Blocking Transformation Effects

A blocking transformation requires the complete set of records cached from the input before it can release records downstream. Earlier in the Understanding and Tuning the Data Flow Engine Topic , you saw a list of nearly a dozen transformations that meet this criterion. The most common examples are the Sort and Aggregate Transformations.

Blocking transformations are intensive because they require caching all the upstream input data, and they also may require more intensive processor usage based on their functionality. When not enough RAM is available in the system, the blocking transformations may also require temporary disk storage. You need to be aware of these limitations when you try to optimize a Data Flow. Mentioning the limitations of blocking transformations is not meant to minimize their usefulness. Rather it’s to emphasize how important it is to understand when to use these transformations as opposed to alternative approaches and to know the resource impact they may have.

Because sorting data is a common requirement, one optimization technique is valuable to mention. Source data that can be sorted in the component through an ORDER BY statement (if the right indexes are in the source table) or presorted in a flat file does not require the use of a Sort Transformation. As long as the data is physically sorted in the right order when entering the Data Flow, the Source component can be configured to indicate that the data is sorted and which columns are sorted in what order. As mentioned earlier, the IsSorted advanced property of the Source component is documented in Joining Data Topic.

Troubleshooting Data Flow Performance Bottlenecks

The pipeline execution reports are a great way to identify which component in your Data Flow is causing a bottleneck. Another way to troubleshoot Data Flow performance is to isolate transformations and sources by themselves.

While you are developing your package, you can identify bottlenecks within a specific Data Flow by making a copy of your Data Flow and begin decomposing it by replacing components with placeholder transformations. In other words, take a copy of your Data Flow and run it without any changes. This will give you a baseline of the execution time of the package.

Next, remove all the Destination components, and replace them with Multicast Transformations (the Multicast is a great placeholder transformation because it can act as a destination without any outputs and has no overhead). below screen shot represents a modified package in which the Destination components have been replaced with Multicast Transformations.

represents a modified package in which the Destination components have been replaced with Multicast Transformations

Run this modified Data Flow and evaluate your execution time. Is it a lot faster? If so, you have identified your problem — it’s one or more of your destinations.

If your package without the destinations still runs the same, then your performance bottleneck is a source or one of the transformations. The next most common issue is the source, so this time, delete all your transformations and replace them with Multicast Transformations, as below screen shot shows.

transformations and replace them with Multicast Transformations

Run your package. If the execution time is just as slow as the first run, then you can be sure that the performance issue is one or more of the sources. If the performance is a lot faster, then you have a performance issue with one of the transformations. Even the Active Time report shown in Figure 16-17 can give deceptive results if the source is the bottleneck. The source and all transformations in the first execution tree will show high active time. In fact, all your Data Flow components may have a high active time if you don’t have any blocking transformations. Therefore, checking the sources will assist in identifying whether the Source component is the hidden bottleneck. 

In fact, the Multicast approach can be applied repeatedly until you figure out where the issue lies. In other words, go back to your original copy of the Data Flow and replace transformations until you have identified the transformation that causes the biggest slowdown. It may be the case that you have more than one component that is the culprit, but with this approach you will know where to focus your redesign or reconfiguration efforts. 

0 Responses on SSIS Data Flow Design and Tuning"

Leave a Message

Your email address will not be published. Required fields are marked *

Copy Rights Reserved © Mindmajix.com All rights reserved. Disclaimer.