When is the BEST time for a Data Quality Review? | Roshan Joseph (via LinkedIn)

Follow the LinkedIn discussion

My comment

While position 5 (NOW!) is the "correct" answer, I like to add "Merger & Acquisition" as a triggering event (variation / combination of pos. 1 to 4).

With an upcoming M&A transaction, a data quality review prepares for the audit that is an indispensable part of the due diligence. Both (all) involved organizations should undergo a data quality review to especially know about the mergability of the parties' data before taking the final decision.

Tool to Track Which Databases Keep Customer Data | LinkedIn Group: Master Data Management Pros

Follow the LinkedIn discussion

My comment

I suggest you to use a professional data and process modeling tool suite.

The data modeling tool component will allow you to have an inventory of the data, i.e. which fields (particularly: customer data) reside in which database. Typical use of the data modeling tool could be:

• Reverse engineer each database, i.e. automatic transfer of the database structure to a graphical/textual representation in the data modeling tool.
• (Since even (semantically) same fields will have different physical names in different databases...) Link synonyms to a common business name, e.g. "cust_name" and "cli_nam" could both represent "customer name". 
• Add/modify any other crucial description that may be missing/incorrect.
• Integrate the database models into subject areas (A subject area will give you the synchronized business view of how e.g. a customer is - and perspectively should be - described in your organization, e.g. by customer first-name, customer family-name, customer date-of-birth, etc.)

The process modeling tool component will provide you a graphical/textual representation how the database fields "flow" through your organization, i.e. which fields are included in the "input" and/or "output" data flow(s) of a process (program, module, dialog,...).

Ideally, the tool suite will be integrated, i.e. database fields that are captured in the database reverse engineering step (using the data modeling tool) can be linked to the fields found in the analysis of the data flows (using the process modeling tool) and vice versa.

How you apply the modeling tool suite in detail will certainly depend on the mid and long-term goals of your organization, e.g. merging/replacing application systems, evaluating new software packages, changing platforms, going mobile etc.

Considering any of these targets combined with your initial question, I recommend you to check out the SILVERRUN Professional & Enterprise Series at www.silverrun.com . (In the spirit of full disclosure: I represent Grandite, the maker of the SILVERRUN tools.)

Please do not hesitate to contact me for further information directly, you will find my coordinates in "Contact Info" of my LinkedIn profile.

How Does the Database Influence the Data Modeling Approach? | LinkedIn Group: Data Modeling

Follow the LinkedIn discussion

My comment

Based on the 4 primary steps suggested by Rémy [Fannader] (even considering that each of us may have slightly different convictions how to exactly mark off these steps against each other), the answer should be:

• Conceptual: not influenced by the target database 
• Normalized logical: not influenced by the target database
• Denormalized logical: only influenced by the architecture / type of concepts that the database supports (doesn't support), e.g. nested table, materialized view
• Physical: completely influenced by the target database, e.g. physical names, database-specific storage parameters.

On an additional note: To follow the above 4-level procedure effectively and efficiently, it is indispensable to use a professional data modeling tool that not only, but particularly allows to

• Define, keep and maintain the above development levels
• Propagate (cascade) applicable modifications to the next level(s)
• Generate the DDL from the physical level. 

How to Identify Parent / Child Role of an Entity in a Data Model Diagram Using "Information Engineering" Notation | LinkedIn Group: Data Modeling

Follow the LinkedIn discussion

My comment

Presuming that many-to-many relationships have been resolved as (binary) one-to-many relationships, there are two ways to communicate / express relationships, as 

• One-to-many relationships (or optional-one-to-one)

and/or

• Parent-child relationships [Child entity is the side of the relationship where the foreign key (constraint) will be added.]

However, provided that the integrity of the model has been positively verified, these two ways are synchronized, i.e. the role of the parent entity and of the child entity in a given one-to-many (or optional-one-to-mandatory-one) relationship can be derived following the rule "a mother can have many children, but a child has a maximum of one mother". 

The exception to this rule is an optional-one-to-optional-one relationship which needs further specification about the parent or child role of the participating entities.

For this latter case (or an interim state where the integrity of the model has not been verified yet), an Entity-Relationship diagram in Information Engineering notation only expresses the multiplicities of a relationship, but does not offer any "standard" indication about the parent / child role of an entity in a relationship.

Therefore, I suggest to use a data modeling tool that allows you to e.g. additionally display the name of the child direction close to the respective side of the relationship connector.

Once the model is verified and ready to generate foreign keys, the latter ones will graphically identify the child role of an entity in a relationship.

Where Does the Line Between Data Modeler and DBA Fall? | LinkedIn Group: InfoAdvisors Members

Follow the LinkedIn discussion

My comment 

I agree with your [Karen Lopez] "State of the Union" of Data Modeling as described in your post related to this discussion. Leaving possible and probable reasons aside that led to that state, I will here focus on your question.

According to my observations, organizations have lost (and some never adopted) the technique of targeted denormalization which bridges the gap between the logical data model and the physical database design and thus brings Data Modelers and DBAs together.

Based on a 3-level approach (I skip to discuss the steps that lead to a Logical Data Model), responsibilities of Data Modelers, Developers and DBAs can be assigned as follows:

• Logical Data Model: developed by Data Modelers / Data Administrators

• Transition Model: created by Developers and DBAs to denormalize the Logical Data Model according to the requirements of the application (create physical tables, views, indexes etc.)

• Physical Database Model: based on the Transition Model, DBAs add physical parameters allowing to completely generate the DDL for the (production) database

To be consequently followed by Data Modelers, Developers and DBAs, the above 3-level approach does not only need to be advised and backed by senior management, but it requires a data modeling tool that will make it "unattractive" for anyone to seek "practical shortcuts". To be suitable for that approach, the data modeling tool e.g. has to

• Support denormalization of Logical Data Models

• Keep track of the lineage through the transformation process from logical tables/columns to database tables/columns

• Offer mechanisms that automatically propagate modifications from the Logical Data Model to subsequent levels (for modifications where the methodology is algorithmic and does not require human intervention)

• Offer mechanisms that allow to manually integrate modifications from the Logical Data Model to subsequent levels (for modifications where design decisions need to be taken)

• Include an interface that generates the script to create / alter the database (DDL) from the Physical Database Model.

How to model a ternary associative entity with a binary constraint? | LinkedIn Group: Data Modeling

Follow the LinkedIn discussion

My comment 

Based on the information provided (and including your later remark that this is just a hypothetical conceptual model to help illustrate a concept), Option A shows the right model.

Additional relationships as of your options B and C are redundant since they do not add any semantics to the model that are not already expressed in Option A. Practical test: If you generate foreign keys, the model as of Option A will look as shown at http://www.silverrun.com/common/ternary-example-red.gif (using the notation "Information Engineering +"). 

Click on image to enlarge it

The foreign keys in the table Diploma still allow a direct navigation to the table University and the table Degree (and vice versa).

A Master Data Mind Map | LinkedIn Group: DAMA International

Follow the LinkedIn discussion

My comment 

Starting with a mind map is definitely a more "relaxed" technique during the brainstorming stage of an MDM endeavor, compared to immediately using a data modeling tool. The relationships between master data entities are simply specializations, so, at the first stage, the creative process is not overloaded with the pressure to name relationships and assign cardinalities.

Therefore, this approach is more suitable to obtain acceptance from the target audience (stakeholders / representatives of the business units). Their engagement and contributions are not only indispensable to find and define master data entities as the center of operational transactions, but also to build a sustainable basis for analytics processes: As you mentioned in your blog, master data are about the Who (Party), What (Product / Service) and Where (Location), i.e. these master data entity categories (together with When = Time) also define the dimensions of the analytics space, as most business questions can be projected into and then answered from this 4-dimensional structure.

Modeling of un-structured data | LinkedIn Group: Data Architect USA

Follow the LinkedIn discussion

My comment 

All business-relevant data should be modeled in only one tool to ensure the integrity between logical enterprise models, logical subject area models and application-specific DBMS models.

Differences between modeling techniques for SQL databases, NoSQL databases and other storage / retrieval technologies only occur on the physical level. In a nutshell, while SQL database models 'may' be denormalized, NoSQL database models 'must' (almost always) be denormalized (due to rare to no support of table joins).

In any event, professional data modeling tools (should) offer such denormalization support while keeping track of the column lineage.

The Business Architecture tool SILVERRUN in its latest version (published last week) features modeling Cassandra 2.0 databases incl. the generation of CQL scripts (see http://www.silverrun.com/silverrun-news-nosql-cassandra-modeling.html ) Additional SILVERRUN versions including reverse engineering of Cassandra databases as well as support of other NoSQL databases will follow in 2014.

[In the spirit of full disclosure: I am in charge of Grandite, the SILVERRUN supplier, and will be happy to provide additional information on demand.]