Tod means Fox

Supporting decisions through sound data management

Archive for category Data Architecture

Mindmapping Dimensional Models

Aug 22

Posted by Tod McKenna in Data Architecture

When presented with a data modeling problem, I start with a conceptual design and then create the logical and physical designs as each concept becomes more mature and stable. This is an iterative process that can take many hours. Using mind mapping software has always given me a head start.

Mind mapping has been around for a long time. It’s a visual technique that you can employ which allows you to diagram ideas based on some central theme. For dimensional modeling, the theme is some event in the business process, while the ideas are the dimensions and dimension hierarchies. Mind maps are quick to make, easy to follow and share, and will allow you to see all interconnected concepts in one place. Software developers and data modelers have been using mind maps for a long time, but its use (as far as I’ve seen) isn’t quite mainstream in the dimensional modeling space.

Orders 300x97 Mindmapping Dimensional Models

Mind mapping an Orders business process dimensional model

When I start to construct a new data model (ER/DM), my first attempt at a design is often a mind map. As I read through requirements and examine business processes and business entities, I start to draw out how they may relate. Traditionally I used a paper and pencil. But recently, I’ve switched to using mind mapping software that I can access and share on all my devices. MinDgo, for example, works on my Mac, iPhone, and iPad. Once I’ve sufficiently covered all key concepts and requirements, I complete the conceptual model in PowerDesigner.

I use mind mapping software for the following reasons:

  1. Quick, easy, and structured way of designing high-level business process dimensional models
  2. Available on all my devices (unlike the heavy case/modeling tools we use), so when inspiration strikes, or when that coffee-machine meeting concludes, I can quickly get the ideas into the design
  3. Very easy to show, explain, and help interpret the models to business and technology colleagues
  4. Organizing different iterations of a design, and interconnecting related designs, is easy using software (try this in your Moleskin)

I also use this technique when I am trying to understand existing data models. For example, if I’m analyzing complex database documentation (from vendors like WorldScope or Charles River), I can get a good feel for how things relate by mind mapping as I go.

Tags: , , , , ,

4 Comments

Chaos Theory and the Data Warehouse

Apr 18

Posted by Tod McKenna in Data Architecture

Have you ever considered the Data Warehouse as a chaotic system? The work of the Data Warehouse team is never complete: new requirements trickle in every day, and user feedback gets more and more sophisticated as time passes. Chaos Theory can help explain this, and in the end, offer us some insight into how we can better plan Data Warehouse development, deployment, and maintenance.

butterfly effect 150x150 Chaos Theory and the Data WarehouseThe Data Warehouse is a process which forms the center of an information supply supply chain, with several inputs and several outputs. Each input and each output is subject to change based on factors such as vendor upgrades, new interfaces, expanded interfaces, and perhaps most importantly end-user (client) evolution. All of these changes happen continuously. As people use the Data Warehouse, they become more inquisitive. They want their output and analysis rolled up or down in different ways. Predicting (i.e. planning) for Data Warehouse change can be as difficult as predicting (and therefore planning for) the weather. This environment of ever-changing needs fits neatly into the confines of Chaos Theory. But what is chaos in this context? What is Chaos Theory exactly?

From the book “Chaos Theory Tamed”, author Garnett P. Williams writes:

Chaos is sustained and disorderly-looking long-term evolution that satisfies certain mathematical criteria and that occurs in a deterministic non-linear system. Chaos theory is the principles and mathematical operations underlining chaos. (pg 9)

Meteorologist Edward Lorenz in the 1960s determined that even the tiniest differences in an initial measurement can have a huge impact on an outcome. In other words, as his butterfly effect posits, a butterfly flapping its wings in Africa can affect weather patterns in North America. Weather is a system which has a highly sensitive dependence on its initial inputs.

The foundation of the Data Warehouse is only as stable as how you control for the tiniest changes to the inputs into the information structure. As weather, it too has a highly sensitive dependence on inputs. One tiny change to a source system can have almost catastrophic effects on the Data Warehouse.

Finding Order

However, despite the chaos, we should be able to find some order. This is what Lorenz and scientists after him tried to do. The first step in this process is understanding that even seemingly random changes are not always as random as they seem. If we can understand that changes to our Data Warehouse are not random, then we can build a better Data Warehouse.

There are a few things you can do to tame the chaos:

  • Be consistent and systematic. The more predictable you and your Data Warehouse team are, the easier it will be handle change. In other words, control any and all variables that you can.
  • Adopt proven analysis and development methodologies that others have had success with. This is not to say that some level of adaptation to your environment, team skills, and situation are not required, but rather, start off with a good foundation and follow along where it makes sense.
  • Keep the team close. Quality and frequent interaction among the people who make and run the DWH is essential.
  • Stay in the groove like an improvisational jazz band. If your data modelers are not in tune with your decision-support analysts who are not in tune with your DBA, then you can’t expect to handle the challenges of chaos.
  • Feedback and evolution are two very important aspects of Data Warehousing. Keep your ear to the wall and try to anticipate changes before they occur. This takes practice, but (back to the improvisational jazz band analogy) practice makes perfect.
  • Keep in step. In the Data Warehouse world, change is natural and will come in waves. More significantly, if changes cannot be implemented quickly, your clients will lose confidence in your ability to keep up.
  • Think and act quickly. The longer you debate, the longer your client must wait. While they wait, they construct workarounds or look elsewhere. If you’re lucky and they do wait for you, their change may become outdated and no longer relevant; an opportunity might have been missed (and you’ve essentially failed them).
  • Don’t be afraid to be wrong. The consequence of acting quickly is that you might get something wrong. Just be agile enough to respond and deliver new change with urgency.

I’ll post more thoughts on this over the next weeks. I’m particularly interested in how users of the Data Warehouse become more and more sophisticated as they use its tools and applications.

Tags: , , , , , ,

No Comments

Data Warehousing ETL Checklist

Aug 27

Posted by Tod McKenna in Data Architecture

IT Business Edge writer Michael Stevens recently published a document called the “Data Warehousing ETL Checklist” (link to article with document). To get the file you’ll need to register with ITBusinessEdge (if you don’t already have an account).

The document isn’t a development checklist, but rather “a high-level checklist of the important topics”. Stevens has made a good effort to distill the various data integration topics and issues you’ll need to consider before you start your project.

Highlights and feedback

Mind scope and do not “try to boil the ocean”. Good advice. But I disagree with Stevens on how to approach one aspect of the scoping issue. He suggests that “in some cases, it may be better for the business as a whole to leave certain data sources untouched”. As I wrote here, I don’t think that scoping to the data source is a great approach. But you can decide.

I would also add that instead of looking at “Target System Content”, you should focus specifically on Business Processes — and this is regardless of the type of system your integrating into (MDM, DW, OLAP, etc.).

Data Profiling is a very important topic, and I’m ecstatic to see it on the list. But if it were up to me, I would have made it second or third. Almost every single failed or over budget ETL project can be linked to data source woes (quality, data availability, inadequate documentation, etc.). Most of these woes can be relieved by conducting good data profiles — a process that can begin during the initial requirement gathering phase of the lifecycle.

The data profile will reveal which source systems are useful. There may be redundancy, data quality, and data access issues compelling you to exclude a particular source. This isn’t a scoping issue, though. Go back to the project sponsor and to the board; report your findings. You may have other issues in your organization to resolve first.

In addition to Data profiling, I’m very happy to see that Naming Conventions made the list. Tied in closely with naming is the concept of conformity. I’ve written about this before, so please check here. In general, a label and its data must conform and be consistent throughout the entire system. And as Stevens points out, they must be agreed upon by the users.

For metadata, I also consider a third type: Process Metadata. This is technical data about all the processes (ETL, jobs, automated processes, reports, etc.). This data is then utilized directly by business users and administrators to monitor and audit the data warehouse. Is my report ready? consult the process metadata. Did that job finish? consult the process metadata.

Lastly, don’t forget to think about auditing, compliance, and lineage. If you’re making decisions based on the data retrieved from your data warehouse, you’ll need ways to trace data lineage. You’ll need to know when data was loaded, what version of the ETL system and business logic was used, and even why a particular data item might have been changed in transit.

Conclusion

This is a good high level checklist that you can use before you begin to dig deeper. It gives you a snapshot of some of the high-level decisions you’ll need to make before you get started with your next data warehousing project.

Perhaps Stevens will incorporate some of my feedback into a version 2!

Tags: , , , , , , ,

No Comments

10 Commandments of Data Integration

Jun 30

Posted by Tod McKenna in Data Architecture

  1. You shall compile and document all requirements and mappings; segregate the work by business process. You may have more than one of these business processes, some of which may come before others.
  2. Do not begin without first conducting a thorough data profile; otherwise, you will be punished for your inequities, as will the generations that come after you.
  3. Do not think commandments one or two are in vain, lest you will become overrun by the dead line, scope creepers, and a great exodus of people from your tribe; if this happens to you, do not swear or curse, for you have been warned.
  4. Remember that latency and timeliness are equal in importance to non-volatility and having a traceable lineage; a staging area may lead you to this promised land.
  5. Honor the rules of data conformance.
  6. Do not kill dirty data: you shall clean them, or take them back to their sources for retribution.
  7. Do not commit the worst data integration transgression of all and ignore data quality, your ignorance will not be forgiven.
  8. Do not be shy about stealing your neighbor’s work, for his trials have led to best practices that you can make equally good use of.
  9. Do not rely solely on business keys; surrogates are your friend and will permit you to engage in slowly changing your dimensions.
  10. You shall covet a proper audit and log system; for on the day of judgment, you will need proof of your compliance.

Tags: , , , , ,

No Comments

ETL Subsystem 31: Paralleling and Pipelining

May 18

Posted by Tod McKenna in Data Architecture

This article is part of a series discussing the Kimball Group’s “34 Subsystems of ETL“. The Subsystems are a group of “Best Practices” for delivering a BI/DW solution. In my articles, I discuss how each Subsystem can be implemented in SSIS or hand coded in Visual FoxPro.

intro 150x150 ETL Subsystem 31: Paralleling and PipeliningOf all the subsystems that I’ve discussed so far, this one resulted in the most research. I had to (a) learn more about how paralleling works and (b) experiment with my environment to better understand it.

Honestly, I’ve taken this subsystem for granted over the years. And for VFP, I’ve done little exploration in this arena. For SSIS, I have tended to adjust the settings I can adjust (more on this below), monitor the results, and tweak my performance as needed. In some environments, this lackluster approach will get you by just fine. If you have very small load windows and performance is critical, then you’ll need to make a better effort.

So what is Paralleling and Pipelining?

Lumped together into a single subsystem, these two performance means are related but different. They’re cousins, I suppose. Running your ETL processes in parallel means that your ETL system is carrying out multiple operations simultaneously. Pipelining your ETL processes means that you can start new operations before the previous ones complete.

Paralleling and Pipelining are quite desirable. And depending on your tool of choice, taking advantage of them can be painless or painful.

How can you apply them?

You can achieve parallel processing by utilizing the CPUs on a single machine, or you can utilize multiple machines. The first option is the easiest to set up and results can be quite good. For a multiple CPU (or core) setup, you are actually running code (programs, algorithms) simultaneously, potentially doubling performance, all on the same box. You can scale out (i.e. scale horizontally) your ETL processes by adding computers (see What is distributed computing? by Kirk Pearson), allowing you to take advantage of the CPUs, RAM, and I/O of each machine. The latter has some significant design implications, but well worth it if your environment needs it.

assembly line women.thumbnail ETL Subsystem 31: Paralleling and PipeliningPipelining increases throughput. Unlike parallelism, it will not allow the instruction to run faster, but rather it permits downstream processes to start before the upstream process finishes. A great analogy is an assembly line, where parts are added to the whole as it travels down the line.

Getting parallelism and pipelining to work together is the Holy Grail of ETL performance. While certain performance techniques are available at all phases of data integration (from Extraction and CDC, to surrogate key handling and using partitions for fast loading), none can compare to the gains you can realize with this subsystem.

You should also keep in mind that CPU multitasking is different from parallel processing, and multithreading is different from pipelining. A multitasking process shares CPU resources, giving the illusion of parallelism (although one man’s illusion can be another man’s reality). Multithreaded applications share the same memory, but operate on different engine threads (i.e. a subtask). Multitasking and multithreading, like pipelining, increase throughput and also play an important role in performance tuning. I’ll talk a little more about this below in my section about FoxPro. Otherwise, if anyone is interested, I can try to elaborate in another post.

Where can this best be utilized?

Here are some ideas on where you can focus your efforts:

  • When loading historical data or retrieving data from multiple similar sources, execute the same package for different date ranges, at the same time (in SSIS for example, use multiple Execute Package Tasks or run the same package multiple times together as Jamie Thomson explores here), or you could design your historical load packages to break apart the data into separate threads.
  • Spread out UPDATE statements. This can be real handy if you have a few accumulating snapshot fact tables.
  • Spread out complex algorithms and routines that can operate on a subset of data.
  • Load staging tables while downstream processing loads your dimensional model
  • Do lookups (especially surrogate key lookups) in parallel
  • Distribute your conformed dimensions to other machines, data marts, etc. in parallel

 

SQL Server 2005 Integration Services (SSIS)

As you know, I use SSIS and VFP for ETL (not at the same time or on the same project though). With SSIS, I can quickly create complex routines that can automatically take advantage of multiple processors. The native support for buffers, execution trees, and parallelism makes my job pretty easy (which is why I suppose I’ve taken this subsystem for granted over the years). Simply understand how SSIS works, adjust the settings you need to adjust, monitor your performance, and tweak as needed.

To get a grip, the following resources are invaluable:

 

Hand Coding with Visual FoxPro (VFP9)

While SSIS and SQL Server have built-in mechanisms to manage most of the paralleling and pipelining responsibilities for you, FoxPro does not. You can achieve some very good results using VFP and multithreading, but you have to be extremely creative in how you handle paralleling and pipelining. If you don’t think this is the case, I’d love to hear how paralleling and pipelining can be achieved with VFP!!!

Of course, the VFP community is — and has always been — quite creative. As with most of this sort of thing, Calvin Hsia is near the front of the line. MTmyVFP (True VFP multi-threading) on CodePlex is a creative example using Hsia’s Multithreading class. For more information and a ton of details, check out:

As I’ve stated before, multithreading is not parallelism, nor is it pipelining. But if you utilize MTmyVFP (or similar solution) in your VFP ETL system, you will realize many performance benefits. Lastly, there was a pretty interesting, albiet short, discussion on this issue here.

From here

This post might have come off a bit long-winded, but there were quite a few important points to make. I hope that I’ve been able to distill what I’ve learned and that in the end, it all makes some sense. In my next ETL post, I’ll talk about ETL Subsystem 32: Security.

Tags: , , , , , , , , , ,

No Comments

ETL Subsystem 30: Problem Escalation

Apr 28

Posted by Tod McKenna in Data Architecture

This article is part of a series discussing the Kimball Group’s “34 Subsystems of ETL“. The Subsystems are a group of “Best Practices” for delivering a BI/DW solution. In my articles, I discuss how each Subsystem can be implemented in SSIS or hand coded in Visual FoxPro.

escalator.thumbnail ETL Subsystem 30: Problem EscalationProblem escalation in a data integration project is much like problem escalation for any deployed application. The primary difference is that most of the escalations are initiated by software and not people.

A typical escalation scenario would start with an end-user or business user report or complaint. In a data integration environment, your data monitoring tools, scheduler, and ETL system will be watching for the exceptional events and states that cannot be handled automatically. These exceptions are then forwarded through your various levels of support (or simply to your data warehouse maintenance group) by email or through some dashboard/support desk application. This process is in addition to calls made by your end users to your help desk.

The ETL Subsytems, remember, are a set of best-practices identified by the Kimball Group for data integration. So it follows that problem escalation and resolution would be an important element. Note that problem escalation is also a major component of your Service Level Agreement (SLA) — A contract between you and your business users which states how you will provide your (DW/BI) service over some period.

The ultimate goal of this Subsystem is to create a highly automated support center that will keep your data integration processes healthy. The support center’s foundation is its escalation plan; in other words, the pathway that an incident takes through your team. This pathway includes stops along the different support levels. Each level has certain capabilities and expertise that can help solve problems as fast as possible.

What are incidents?

An incident, according to ITIL, is “any event which is not part of the standard operation of a service and which causes, or may cause, an interruption to, or a reduction in, the quality of that service.” They range from end-user complaints to server crashes. Problems, issues, and complaints are all types of incidents.

I like to categorize ETL incidents into 3 broad groups:

  • Data (where quality, latency, or reliability is the primary issue)
  • Process (where one of the ETL components is failing due to some exception)
  • Infrastructure (where the network, hardware, middleware, or any supporting software fails)

After the incident is categorized, a severity level can be assigned.

Severity

I recommend taking the time to develop a good severity matrix. A severity matrix is a nothing more than a table with the following headings: Severity Level and Description, Response Time, and Resolution Time. The matrix will help you determine what incidents are sent where, how responses on the issues should be met, and what the expected turnaround times should be.

ETL Subsystem 30: Problem Escalation

The above image is an example of a severity matrix taken from Information Security Short Takes, in an article titled “9 Things to watch out for in an SLA“. A very good read if you want to know more about SLAs!

Severity Level and Description
Usually you will see severity levels ranging from 1 to 4. The rankings depend entirely on your organization, your IT framework (if you use ITIL, for example), and the range of users using the data warehouse. As an example:

  1. Critical: A level reserved for situations when your data warehouse or BI applications are non-functioning
  2. High: Any non-critical issue that prevents one or more people from doing their job
  3. Medium: All other problems not deemed high or critical
  4. Normal: User requests, such as a new installation, that are not deemed to be true incidents or perhaps known issues that will be addressed in some future release
Response Time
Once an incident is reported, what is the expected feedback time? For critical issues, the feedback should be immediate. For Normal requests, the requesting party or parties should be notified as soon as possible and practical.
Resolution Time
This is the expected time it takes to resolve an issue. Critical issues must be resolved quickly and could involve all of your resources, while simple requests would be handled as time and resources permit.

The next step is defining your escalation groups — the people in charge of handling the various incidents.

Escalation Groups

Remember that ITIL and other frameworks already define how your escalation groups should look and interact. But also keep in mind that data integration is much different than typical applications. Your customers are generally high-level analysts and power users, managers who are responsible for P&L, and all those wonderful C-Level executives who expect this heavy Business Intelligence (or SOA, MDM, etc.) investment to run flawlessly all the time.

If you are not operating under an IT framework, and you have some flexibility in how you handle incidents, then consider the following:

  • Create 3 Escalation Groups, or “lines of support”:
    1. Triage – will organize and distribute incidents appropriately (i.e. the help desk)
    2. Analysts and your Data Steward – will be responsible for thinking through and building resolution plans for data and process problems
    3. DBAs, network admins, and the development team – will do the work required to correct the issue

  • Automate almost every part of the incident reporting process. This avoids interaction with Triage, saving precious time. An added benefit is that some issues can be resolved before a manager has to make a call. Some examples:
    • Send an email to the support team if a job fails, a report crashes, or some other process is interrupted
    • Pick up a 3rd-party bug report system that can be installed on the company intranet and allow your users to access and post to this system
    • Infrastructure issues should go directly to the IT department in charge of the component, skipping the need for level 2 support.

  • Do not outsource critical support personnel. Unless the contractors are engaged with your organization and business, they likely won’t care enough to truly “own” the problem (unfortunately, this is the case where I work: the outsourced help just doesn’t “get it” and often lazily handle issues business users deem critical).

From here

As an ETL developer and/or architect, problem escalation might not be on your radar. But it should be. Everything you build should self-report when an exception occurs. That’s the key to automating this Subsystem. You can’t do it afterward (at least not easily), so it must be in the initial planning.

In my next post, I’ll dive into ETL Subsystem 31: Paralleling and Pipelining.

 

Tags: , , , ,

No Comments

SQL Server Deployment and Performance Tips and Advice

Apr 19

Posted by Tod McKenna in Data Architecture

In line with my previous post, I wanted to share with you an instructional video I found some time ago on YouTube. Kevin Kline (author of SQL In A Nutshell) walks through several important topics and challenges in deploying and maintaining good performance for SQL Server. Although he uses SQL Server 2005, most (if not all) of this information applies to SQL Server 2008 as well.

It’s a bit long, but well worth it.

 

Tags: , ,

1 Comment