Historical content within data structures increases their complexity. Making complexity "easy" for the initial developers and "harder" for the later users is a short-sighted approach.
Posted March 12, 2014
In many organizations, users find it hard to trust their own internal information technology (IT) group, leading them to try any possible option to solve problems own their own. The resulting stealth IT projects can lead to confusion or even complete chaos.
Posted February 10, 2014
Changes and enhancement to solutions are hard, even under the best of circumstances. It is not usual that, as operational changes roll out into production, the business intelligence area is left uninformed, suggesting that data warehouses and business intelligence be categorized according to the view of the old comedian Rodney Dangerfield because they both "get no respect."
Posted January 07, 2014
In the world of database products, trends seem to pop up quickly, like prairie dogs testing the air. However, regardless of the marketing jargon, or platform, vendor-product purchased, or open source utility downloaded, the one thing that remains unaltered is that in order to extract value from data, the data must be understood. The individuals within any organization who actually comprehend the data, the data structures, and all the exceptions to the usual rules are individuals who are considered critical resources.
Posted December 04, 2013
Changes to database structures should be performed in a coordinated fashion as the application processes that support the new functionality are rolled out into production. While the "work" involved in adding a column or a table to a relational database is actually minimal, often there are circumstances where developers and DBAs create additional columns and additional tables in anticipation of future needs. Sadly, this "proactive" effort results in databases littered with half-formed ideas, fits-and-starts, and scattered-about columns and tables that provide no meaningful content.
Posted November 13, 2013
How does one avoid the semantically wishy-washy use of NULL-surrogates and instead, actually design structures wherein NULLs are not necessary?
Posted October 09, 2013
Database management systems support numerous unique date and time functions - and while the date-related functions are many, they do not go far enough. One date-driven circumstance often encountered has to do with objects having a type of date range that needs to be associated with it. While there are some exceptions, this date range need generally ends up implemented via two distinct date columns—one signaling the "start" and the other designating the "end." Maybe, should the creative juices of DBMS builders' flow, such things as numeric-range-datatypes could be created in addition to a date-range data-type. Who knows where things could end up?
Posted September 11, 2013
Data models attempt to express the business rules of an organization. A good data model reflects the semantics used within an organization to such an extent that business people within that organization can relate to and easily agree with what is being expressed. In this regard the data modeler's goal is to properly mirror back the organization's concepts onto those people within the organization. The goal is not to force an organization into a "standard" data model, nor is the goal to abstract everything in the creation of a master model that will never need to change even if the business rules were drastically shifted.
Posted August 07, 2013
One of the principles within relational theory is that each entity's row or tuple be uniquely identifiable. This means the defined structure includes some combination of attributes whose populated values serve to identify an individual row within the table/relation. This, or these, attribute(s) are the candidate key(s) for the structure. The candidate key is also known as the primary key, or if a structure has multiple candidate keys, then one of them is designated as the primary key. When building up a logical design, primary keys should be identified by the actual data points in play.
Posted July 09, 2013
The grain of a fact table is derived by the dimensions with which the fact is associated. For example, should a fact have associations with a Day dimension, a Location dimension, a Customer dimension, and a Product dimension, then the usual assumption would be for the fact to be described as being at a "by Day," "by Location," "by Customer," "by Product" metrics level. Evidence of this specific level of granularity for the fact table is seen by the primary key of the fact being the composite of the Day dimension key, Location dimension key, Customer dimension key, and Product dimension key. However, this granularity and these relationships are easily disrupted.
Posted June 13, 2013
It seems that juggling is the most useful of all skills when embarking on a data warehousing project. During the discovery and analysis phase, the workload grows insanely large, like some mutant science fiction monster. Pressures to deliver can encourage rampant corner-cutting to move quickly, while the need to provide value urges caution in order not to throw out the proverbial baby with the bath water as the project speeds along. Change data capture is one area that is a glaring example of the necessary juggling and balancing.
Posted May 09, 2013
Dimensions are the workhorses of a multidimensional design. They are used to manage the numeric content being analyzed. It is through the use of dimensions that the metrics can be sliced, diced, drilled-down, filtered and sorted. Many people relate to dimensions by thinking of them as reference tables. Such thoughts aren't exactly accurate. A dimension groups together the textual/descriptor columns within a rationalized business category. Therefore, much of the content coming from relational tables may be sourced from reference tables, but the relationship between each source reference table and the targeted dimension is unlikely to be one-for-one. These grouped-format dimensions often contain one or more hierarchies of related data items used within the OLAP queries supported by the structures.
Posted April 10, 2013
Do not allow well-meaning but confused proponents to obscure concepts related to normalization and dimensional design. Under a normalized approach one usually would not expect for numeric data items and textual data items to fall into different logical relations when connected to the same entity object. Yet within a multidimensional approach that is exactly what happens. Multidimensional design and normal design are not the same, and one should not expect to claim that both approaches were used and that they resulted in the same data model.
Posted March 14, 2013
Establishing a data warehousing or business intelligence environment initiates a process that works its way through the operational applications and data sources across an enterprise. This process focuses not only on identifying the important data elements the business lives and breathes, but the process also tries very hard to provide rationality in explaining these elements to business intelligence users.
Posted February 13, 2013
Multi-dimensional design involves dividing the world into dimensions and facts. However, like many aspects of language, the term "fact" is used in multiple ways. Initially, the term referred to the table structure housing the numeric values for the metrics to be analyzed. But "fact" also is used to refer to the metric values themselves. Therefore, when the unique circumstances arise wherein a fact table is defined that does not contain specific numeric measures, such a structure is referred to by the superficially oxymoronic characterization of a "factless fact."
Posted January 03, 2013
Within the information technology sector, the term architect gets thrown around quite a lot. There are software architects, infrastructure architects, application architects, business intelligence architects, data architects, information architects, and more. It seems as if any area may include someone with an "architect"status. Certainly when laying out plans for a physical building, an architect has a specific meaning and role. But within IT "architect" is used in a much fuzzier manner.
Posted December 06, 2012
In writing a definition for an entity, an attribute, or any other element within a database design, the desired end is a descriptive text that is clear, factual and concise. Semantics are an ambiguous and often painful tool to employ. Balancing the need for clarity against the desire to avoid redundancy can be a juggling act that is hard to accomplish. One might not easily recognize what is complete versus what is lacking, versus what has gone too far. But even so, within a definition if one finds oneself listing valid values and decoding the value's meaning, then one has likely already moved beyond what is "concise." Lists of values easily add bulk and quantity of verbiage into a definition, yet such lists do not usually increase the quality of a definition.
Posted November 13, 2012
The beauty of a truly wonderful database design is its ability to serve many masters. And good database designers are able to empathize with those who will use their designs. In business intelligence settings, three perspectives deserve consideration when composing designs.
Posted October 10, 2012
It seems easy to fall into a state where projects and activities assume such soft-focus that routine takes control, where one simply does necessary tasks automatically, no questions are raised regarding what is moving through the work-life production line and everyone is essentially asleep at the switch. Certainly, we may have one eye open ensuring that within a broad set of parameters all is well, but as long as events are basically coloring inside the borders we continue allowing things to just move along. In this semi-somnambulant state we can easily add columns to tables, or even add new entities and tables, or triggers and procedures to our databases, then eventually at some point down the road have someone turn to us and ask, "Why this?" or, "What does this really mean?" And at that point, we surprise ourselves with the discovery that the only answer we have is that someone else told us it was what we needed, but we do not really understand why it was needed.
Posted September 11, 2012
A database design may occasionally show evidence that it lacks proper prioritization. Data models should express truths about the business, or about the universe of discourse. But in expressing business truth this does not mean a data model should express absolutely every truth that anyone might conceive. Some relationships are significant while other relationships are not. And as a general rule, database design is not an exercise in trivial pursuit. Insignificant truths only clutter up a design, increasing complexity, causing users' eyes to glaze over more quickly, and adding no real value towards the endeavors of the enterprise.
Posted August 09, 2012
The whole world can be divided into two groups, these being splitters and lumpers. Design battles are waged across conference rooms as debates rage over whether to split or to lump. Splitters take a group of items divide them up into sub-groups and sub-sub-groups occasionally going so far as to end with each lowest level becoming a group of one. On the other side of the design fence, lumpers combine items until everything is abstracted into group objects covering very broad territory, such as a "Party" construct, or ultimately an "Object" object. Within data modeling, arguments arise, such as whether to sub-type an entity. Or perhaps lumping is discussed as the grain of a multidimensional fact is proposed. This debate underlies much of the decision-making involved in determining what domains to create within a data model. The split-versus-lump issue is ubiquitous and universal. The question to split or lump arises across many kinds of choices, in addition to the entity definition, table grain, or the domain grain mentioned in the previous examples; this issue is at the heart of deliberations regarding establishing functions, overriding methods, or composing an organizational structure.
Posted July 11, 2012
In the dim, dark past of data warehousing, there was a time when the argument was put forward that "history does not change." It was posited that once a piece of data was received by the data warehouse, it was sacrosanct and nonvolatile. A fact record, once processed, was to remain unchanged forever. Dimensions, due to their descriptive nature, could be changed following the prescribed Type 1, 2, or 3 update strategies, but that was all. It was the expectation that due to their very nature, fact tables would become huge and in being huge would give poor update performance; performance so poor that updates would be virtually impossible to enact.
Posted June 13, 2012
It seems only reasonable that what one person can do, others can learn. On the other hand, taking people through training does not usually result in the creation of great new database administrators (DBAs). It often appears as if those who are exceptional at the craft operate at higher levels as they dive into a problem. Can training alone provide folks with the attention to detail, the urge to keep digging, or the ability to recall minutiae that allow them to rise from simply holding the DBA title to becoming someone who is a great DBA? Or must the genetic potential exist first, and then one might fall into the DBA occupation and astound those around them. It is very hard to say with any degree of certainty whether great DBAs are made or born; yet again the battle between nature and nurture arises.
Posted May 09, 2012
Organizations are populated with solutions entitled DW, EDW, BIC, ODS, EIW, IW, CIF, or BW. Why must every organization have a data warehousing or analytics solution identified by monikers from a very limited pool of choices? Why must every deployment of a database that is expected to function as an operational data store be called ODS? For internal solutions it seems that plain and dreary naming approaches are de rigueur. Two and three letter acronyms have long been a part of corporate-speak; but when it comes to IT systems, these TLAs have become exceedingly narrow and soul-less.
Posted April 11, 2012
Solution development work is usually accomplished via projects, or a combination of programs and projects. This project perspective often leads to thoughts of documentation as project-owned. And while many documents are project-specific, such as timelines, resource plans, and such, not everything is project-specific. Unless projects are established in a fashion whereby each is very limited in scope to the creation or enhancement of a single application or system, specification and design documents belong to the final solution and not to the project.
Posted March 07, 2012
Lectures related to master data bring forth all sorts of taxonomies intended to help clarify master data and its place within an organization. Sliding scales may be presented: at the top, not master data; at the bottom, very much master data; in the middle, increasing degrees of "master data-ness." For the longest of times everyone thought metadata was confusing enough ... oops, we've done it again. And, we have accomplished the establishment of this master data semantic monster in quite a grand fashion.
Posted February 09, 2012
Retaining the particulars of change over time is a fairly intricate configuration. Audit log or shadow tables are sometimes employed, but on occasion there is a need for the "old" and "new" rows to exist in a single operation table for application use. Far too often, the implementation of temporal data structures is shoddy, loose, and imprecise; rather than the fairly complex dance move such temporal arrangements must perform in actuality. The sub-optimal result is much like one's performance of the Funky Chicken at a friend's wedding; the desired moves are mimicked, after a fashion, but it is unlikely to earn high marks on "So You Think You Can Dance." The usual temporal implementation simply slaps on start and stop dates, debates a little over default date values versus NULLs, then moves on to the next subject.
Posted January 11, 2012
The cost for new development can often be easily justified. If a new function is needed, staffing a team to create such functionality and supporting data structures can be quantified and voted up or down by those controlling resources. Money can be found to build those things that move the organization forward; often, the expense may be covered by savings or increased revenue derived from providing the new services.
Posted December 01, 2011
It is not magic. Building a successful IT solution takes time. And that time is used in various ways: obtaining an understanding of the goal; mapping out what components are necessary and how those components interact; testing components and their interaction; and finally migrating those components into the production environment - otherwise known as analysis, design, development, testing, and deployment. Regardless of the methodology employed, these functions must always be addressed. Different approaches focus on differing needs and aspects. But any complete methodology must fill in all the blanks for accomplishing each of these tasks.
Posted November 10, 2011
Every project is, or should be, driven by user requirements. Requirements are the organization's way to articulate what needs to happen in order to provide value. Yet time and again requirements are looked at as something overly technical, mysterious, and too confusing to easily handle. Repeatedly, organizations use a template to ensure that requirements are defined early in the solution process. Sadly, the intended purposes are habitually defeated as these templates are filled with a lack of understanding for what information belongs in a given section, resulting in people creating documents that the authors themselves do not understand. Across many organizations requirements documents are created, reviewed, and even agreed to, that far too often are incoherent monstrosities saying nothing of actual value.
Posted October 15, 2011
When assembling a database design, one of the keys for success is consistency. There should be more than just similarity in the way things are named, the manner in which tables or groups of tables are constructed; the manifestation of these elements should follow standards and practices that are documented and understood. If one tries to rely on the idea that individual developers will simply look at existing tables and glean standards via osmosis as they add on or create new tables, then one does not actually have any standards at all.
Posted September 14, 2011
The Broadway tune goes, "The sun will come out tomorrow ... it's only a day away." The words from this optimistic jingle are often heard on IT projects that are overburdened with features and functions. On any project of significant size the list of desired things often becomes larger than the budgeted resources or time. Faced with limiting circumstances, the only option becomes aligning the work effort with the constraints and only doing what fits within those constraints. The items are "timeboxed," and the amount of planned work is exactly the amount of allowed work. Some things remain in, while other features are left out. Alternatively, responsibility for controlling a project may be ignored and the end date arrives with some things simply not completed. Essentially, a project is timeboxed by default.
Posted August 11, 2011
Naïve approaches to business intelligence will occasionally trap designers as they juggle operational data stores and data warehouses. The trap results from an honest endeavor to simplify designs and increase consistency throughout the solution. Under the umbrella of consistency a designer may plan for a reference table used for operational look-ups to perform a second service as a star schema dimensional table. Some or all reference tables then are declared by fiat to also be dimensions. While on a superficial level there are similarities between dimension tables and more normalized look up or reference tables, fundamentally these two concepts are separate things.
Posted July 07, 2011
Occasionally, one sees a data structure abomination. This atrocity involves an object of almost any type, in almost any database wherein the object has a start date but no end date. It is not that the finish date currently has no value and is null; it is that the end date does not even exist on the table structure. The stop date was never intended to exist. The object in question starts, but it doesn't ever end.
Posted June 08, 2011
Tables within a database management system (DBMS) need primary keys and defined indexes in order for the DBMS to have the opportunity to provide good query performance. Without indexing, the worst possible query performance is guaranteed. The content of these non-keyed tables remains unknown to the DBMS, a black box, where the only possible approach for query execution is to read every row one-by-one. Under such scenarios, the DBMS is little more than a file server that operates more slowly than usual. Without the primary key and indexing, the DBMS may expend even more processor cycles in moving data in and out of the DBMS proprietary storage areas than a file server uses in opening files.
Posted May 12, 2011
Dates are important. Without dates how can anything be planned? However, due dates have been know to increase in importance in the delivery of software solutions. Sometimes the due date becomes such an overwhelming creature of importance that the date is more important than following best practices, more important than verifying that what is built is correct, more important than the solution team gaining a proper understanding of the work they are attempting to perform.
Posted April 05, 2011
The understanding of object states and their transitions obviously is of great importance to the solution developers because as processes are built they will need to support each state and every possible transition. Additionally, knowledge of object states and transitions is of vital importance to the data modeler because the data must be persisted across each of those states, and often the state of an object needs to be easily identifiable. A data model minimally requires a reference table, along with the varying entities that reference that table (the foreign keys tracking an individual object's status). Specific states drive variations of required attributes or combinations of those attributes that apply to one state and not another. The logical definition of the database can identify these variations through the use of supertype/subtype constructs.
Posted March 09, 2011
Referential integrity helps manage data by enforcing validation between related entities. This enforcement follows logical semantics behind the database design -- i.e., an employee can only work for an already defined department; a prescription can only be written by a health care practitioner with the proper authority. A Foreign Key on an Employee table rejects data when any attempt is made to insert or update a row with a department value that does not already exist as a department identifier within a Department table.
Posted February 02, 2011
Back in the 1970s, the ANSI SPARC three-tiered model arose, foreshadowing a smooth intertwining of data and architectural design. The three tiers concept isolated the physical storage needs of data structures independent of business' perception of these structures. The three levels were comprised of schemas labeled external, conceptual, and internal, with each level describing the data in focus from varying perspectives.
Posted January 07, 2011
How does one know what one doesn't know? When evaluating what one knows, it is hard to know where to begin. The wise men say, "The more you know, the more you know you don't know." If one believes such commentary, what is known constitutes the tip of the proverbial iceberg. Databases have an easier time with such missing circumstances. If the rows of a database table are lacking referents, an outer join query filtering for NULLs might detail for you all the missing items. In developing and delivering projects, such a reference list for our minds to link to does not exist, for an outer join or anything else. Often, we do not know everything that needs to be done, particularly as a project starts. The difference between success and failure is not so much what one knows, but in how one handles the gaps between what is known now and what needs to be known before one finishes.
Posted November 30, 2010
Psychiatrists report play is important for keeping our minds and bodies healthy and happy. It has been uncovered that without at least some play, we mentally and physically deteriorate. Lack of play can lead to depression. Our brains are so hardwired in their need for playful activities that without such expressions we loose focus. Play enhances the creativity and imagination within us. Likewise, play has a proper role within our work. And not just for the psychological benefit to the individual, but for the overall health of the organization. For an IT development group, play is essential for finding better ways to approach problems. Within an IT context, play involves evaluating new ways to handle old problems. Prototyping, at times, constitutes play. Prototypes can be used for more than just the practicalities of the user interface; sometimes a prototype of specific functions is necessary for comparisons in order to judge how non-obvious techniques may perform. Such prototype endeavors can involve not only new algorithms, but potentially variant data structures. Ideally, such prototypes will improve team understanding of the data structures and offer insights into potential new and innovative approaches.
Posted November 09, 2010
As bizarre as it seems, no one sabotages our efforts more easily than ourselves. We often leave ourselves vulnerable to failure whenever we respond to requests for project estimates. Like "A Tale of Two Cities" estimating can bring our best and our worst. In offering estimates, individuals may undercut their own efforts. For example, a developer may only consider time coding, and fail to include enough headroom for testing, rework, or documentation. While some aspects of these estimating attempts might suggest a passive-aggressive approach to avoid tasks, much of this estimate-short-sheeting often results from a desire to please and provide a number to make management "happy." There is always a desire from management to have more things done, and to do things quickly on every project. Even when not stated, this unspoken desire colors the responses of the individuals doing the estimating.
Posted October 12, 2010
As databases are established, particularly databases intended to support analytics initiatives, responsibilities for the design must include articulating the planned approaches for enhancing and scaling the database over time. If a database is created to express a multidimensional data warehouse bus architecture, or a corporate information factory, or anything else, the explanation of this connection should exist somewhere. Such documentation should also expand on why things were decided as they were and the expected stylings to be associated with proposed enhancements. Descriptions involving anticipated processing patterns extend naturally from such architecture artifacts. Database and application design personnel should work together in the creation of such credentials to ensure these documents thoroughly cover the needs of the personnel involved in building and maintaining the solution.
Posted September 07, 2010
Design should be an intention, preferably a planned intention. In that intention, design requires more groundwork than a simple thought-train such as the following, "I planned to write a module that functions; since the module functions, my designs are working." Some situations do exist where true design really is less important than successful functionality. Determining the appropriate level of design and preparation offers an interesting question to every architect. At the lowest level, standards and practices present suitable patterns that can serve as a design skeleton for those low-level or isolated items not requiring a heavy-handed blueprint.
Posted August 10, 2010
When integrating data, evaluating objects from multiple sources aids in determining their equivalence. Each source may identify customers, but determining which customer from each system represents the same customer can prove daunting. Sometimes matching things is straight-forward; for example, if all sources should have an accurate social security number or taxpayer ID, success involves simply linking the matching numbers.
Posted July 12, 2010
Quality can be a hard thing to define. What is good and what is bad may not be easily identified and quantified. When a data mart accurately reflects data exactly as found in the source, should that be considered a quality result? If the source data is bad, is the data mart of high quality or not? If the data mart differs from the source, when is the difference an improvement of quality and when is said difference evidence of diminished quality? While it may seem self-evident that correcting the source of load data would be the "right" thing to do, in practice that direction is not necessarily self-evident. The reasons supporting this nonintuitive approach are varied. Sometimes changes to the source impact other processes that must not change, or the changes will expose problems that may provoke undesired political fallout, or it may simply be that making the proper adjustments to the source application would prove too costly to the organization. For all these reasons and more, in the world of business intelligence, the dependent data often is expected to be of higher quality than the source data. In order for that improvement to occur, data placed within the dependent mart or data warehouse must be altered from the source. Sometimes these alterations become codified within the process migrating data from the source. Other times changes are made via one-time ad hoc updates. Either way, this alteration leads to a situation in which the dependent data will no longer equate one-for-one to the source data. Superficial comparisons of this altered content will highlight the disparity that what exists for analytics is not the same as what exists for the operational system.
Posted June 07, 2010
It is rare to find an implemented database devoid of reference tables. Reference tables provide valid values for drop-down lists and in a slightly obtuse way, also allow for expression of a domain or user-defined data type within the database design. Nominally a reference list is but a list of values to be referenced. In practice, the implemented structure of a reference table is driven by small nuances regarding how each list is used. The signifiers may consist of a code or number that creates a short-hand way of expressing a state or value. Yet in addition to such simple codes, a "medium length" name column or even a column to contain a lengthy text description should exist in support of these codes. The names may be used in drop-down lists so that people need not memorize many code values, and the descriptions can assist help-functions by providing users with more understanding of a value's meaning or intended use. For operational applications it may be worthwhile to have an attribute expressing the order in which items should be displayed, unless alphabetical display is a preference. If codes are maintained via online screens, an indicator may be helpful in flagging values that are part of the system and not to be removed or deactivated via those online screens.
Posted May 10, 2010
Far too often we all talk past one another. This cross talking, while not always drastic, remains perceived as an understood fuzziness. Much of the time we ignore these minor miscommunications because precision and clarity are not necessarily critical in all situations. If the general gist is effectively understood between those communicating, that generality may be all that is necessary. Those involved in the communication may feel comfortable that assumptions made to "fill in the gaps" will fall within an acceptable range. Although the lack of clarity in the message communicated may be acceptable, in other circumstances it may not be acceptable.
Posted April 07, 2010
Primary keys come from candidate keys. Each candidate key consists of the attribute or attributes used to label a distinct row in a table. Every candidate key should contain the fewest number of attributes possible to identify rows individually and uniquely. Every entity within a design requires at least one candidate key.
Posted March 04, 2010
Designing a data model that supports the reporting and analytical functions is no different, initially, than any other modeling effort. Understanding the data is crucial. The data architect or modeler needs to feel comfortable with dimensional modeling techniques and needs to obtain a working knowledge of the universe of discourse for the subject-at-hand. A good start in gathering this knowledge begins by reviewing the operational data structures containing the identified source elements. The challenge in designing analytical solutions is found in applying best practices for analytics simply and effectively.
Posted February 09, 2010