Transcript for:
Understanding Database Normalization Techniques

If you’ve had some exposure to relational  databases, you’ve probably come across the term   “normalization”. But what is normalization?  Why do we do it? How do we do it? And what   bad things can happen if we don’t do it? In this video, we’re going to explore database   normalization from a practical perspective.  We’ll keep the jargon to a minimum,   and we’ll use lots of examples as we go. By the  end of it, you’ll understand the so-called normal   forms from First Normal Form all the way up to  Fifth Normal Form – and you’ll have a clear sense   of what we gain by doing normalization,  and what we lose by failing to do it.   This is Decomplexify, bringing a welcome  dose of simplicity to complex topics.   Data: it’s everywhere. And some of it is wrong.   By and large, even a good database  design can’t protect against bad data.   But there are some cases of bad data that a good  database design can protect against. These are   cases where the data is telling us something  that logically cannot possibly be true:   One customer with two dates of birth is  logically impossible. It’s what we might   call a failure of data integrity. The data can’t  be trusted because it disagrees with itself.   When data disagrees with itself, that’s  more than just a problem of bad data.   It’s a problem of bad database design.   Specifically, it’s what happens when a  database design isn’t properly normalized.   So what does normalization mean?  When you normalize a database table,   you structure it in such a way that  can’t express redundant information.   So, for example, in a normalized table, you  wouldn’t be able to give Customer 1001 two dates   of birth even if you wanted to. Very broadly, the  table can only express one version of the truth.   Normalized database tables are not only  protected from contradictory data, they’re also:   easier to understand easier to enhance and extend   protected from insertion  anomalies, update anomalies,   and deletion anomalies (more on these later)   How do we determine whether a table isn’t  normalized enough – in other words, how do   we determine if there’s a danger that redundant  data could creep into the table? Well, it turns   out that there are sets of criteria we can use to  assess the level of danger. These sets of criteria   have names like “first normal form”, “second  normal form”, “third normal form”, and so on.   Think of these normal forms by analogy to safety  assessments. We might imagine an engineer doing a   very basic safety assessment on a bridge. Let’s  say the bridge passes the basic assessment,   which means it achieves “Safety Level  1: Safe for Pedestrian Traffic”.   That gives us some comfort, but suppose we want to  know if cars can safely drive across the bridge?   To answer that question, we need the engineer to  perform an even stricter assessment of the bridge.   Let’s imagine that the engineer goes ahead and  does this stricter assessment, and again the   bridge passes, achieving “Safety Level 2: Safe  for Cars”. If even this doesn’t satisfy us,   we might ask the engineer to assess the bridge  for “Safety Level 3: Safe for Trucks.” And so on.   The normal forms of database  theory work the same way.   If we discover that a table meets the  requirements of first normal form,   that’s a bare minimum safety guarantee. If  we further discover that the table meets   the requirements of second normal form, that’s  an even greater safety guarantee. And so on.   So let’s begin at the beginning,  with First Normal Form.   Suppose you and I are both  confronted by this question:   “Who were the members of the Beatles?” You might answer “John, Paul, George, and Ringo”.   I might answer “Paul, John, Ringo, and George”. Of course, my answer and your answer are   equivalent, despite having the  names in a different order.   When it comes to relational databases, the same  principle applies. Let’s record the names of the   Beatles in a table, and then let’s ask the  database to return those names back to us.   The results will get returned to us in an  arbitrary order. For example, they might   get returned like this. Or like this.   Or in any other order. There is no “right” order. Are there ever situations where there’s a right   order? Suppose we write down the members  of the Beatles from tallest to shortest,   like this. We title our list “Members Of  The Beatles From Tallest To Shortest”.   In this list, it’s not just the names that  convey meaning. The order of the names conveys   meaning too. Paul is the tallest, John is the  second-tallest, and so on. Lists like this are   totally comprehensible to us – but they’re not  normalized. Remember, there’s no such thing as row   order within a relational database table. So here  we have our first violation of First Normal Form.   When we use row order to convey information,  we’re violating First Normal Form.   The solution is very simple. Be explicit – if  we want to capture height information, we should   devote a separate column to it – like this. Or even better, like this.   So far, we’ve seen one way in  which a design can fail to achieve   First Normal Form. But there are others. A second way of violating First Normal Form   involves mixing data types. Suppose our  Beatle_Height dataset looked like this.   If you’re accustomed to spreadsheets, you’ll be  aware that they typically won’t stop you from   having more than one datatype within a single  column – for example, they won’t stop you from   storing both numbers and strings in a column. But  in a relational database, you’re not allowed to be   cagey or ambiguous about a column’s data type.  The values that go in the Height_In_Cm column   can’t be a mix of integers and strings. Once you  define Height_In_Cm as being an integer column,   then every value that goes into that column  will be an integer – no strings, no timestamps,   no data types of any kind other than  integers. So: mixing datatypes within a column   is a violation of First Normal Form, and in fact  the database platform won’t even let you do it.   A third way of violating First Normal Form is by  designing a table without a primary key. A primary   key is a column, or combination of columns,  that uniquely identifies a row in the table.   For example, in the table Beatle_Height,  our intention is that each row should tell   us about one particular Beatle, so we ought to  designate “Beatle” as the primary key of the   Beatle_Height table. The database platform will  need to know about our choice of primary key,   so we’ll want to get the primary key into  the database by doing something like this.   With the primary key in place, the  database platform will prevent multiple   rows for the same Beatle from ever  being inserted. That’s a good thing,   because multiple rows for the same Beatle would  be nonsensical, and perhaps contradictory.   Obviously, a Beatle can’t have  two different heights at once.   Every table we design should have a primary key.  If it doesn’t, it’s not in First Normal Form.   The last way of failing to achieve  First Normal Form involves the notion of   “repeating groups”. Suppose we’re designing  a database for an online multiplayer game.   At a given time, each player has a number  of items of different types, like arrows,   shields, and copper coins. We might  represent the situation like this.   A player’s inventory is what we call a “repeating  group”. Each inventory contains potentially many   different types of items: arrows, shields,  copper coins, and so on; and in fact there   may be hundreds of different types of items  that a player might have in their inventory.   We could design a database table that  represents the Inventory as a string of text:   But this is a terrible design because  there’s no easy way of querying it.   For example, if we want to know which players  currently have more than 10 copper coins,   then having the inventory data  lumped together in a text string   will make it very impractical to write  a query that gives us the answer.   We might be tempted to  represent the data like this.   This lets us record up to 4 items per inventory.  But given that a player can have an inventory   consisting of hundreds of different types of  items, how practical is it going to be to design   a table with hundreds of columns? Even if we were  to go ahead and create a super-wide table to hold   all possible inventory data, querying  it would still be extremely awkward.   The bottom line is that storing a repeating group  of data items on a single row violates First   Normal Form. So what sort of alternative  design would respect First Normal Form?   It would be this. To communicate the fact that   trev73 owns 3 shields, we have a row for Player  “trev73”, Item_Type “shields”, Item_Quantity 3.   To communicate the fact that  trev73 also owns 5 arrows,   we have a row for Player “trev73”, Item_Type  “arrows”, Item_Quantity 5. And so on.   And because each row in the table tells  us about one unique combination of Player   and Item_Type, the primary key is the  combination of Player and Item_Type.   So let’s review what we know  about First Normal Form.   1. using row order to convey  information is not permitted   2. mixing data types within the  same column is not permitted   3. having a table without a  primary key is not permitted   4. repeating groups are not permitted Next up: Second Normal Form.   Let’s look again at our Player Inventory table. This table is fully normalized. But suppose we   enhance the table slightly. Let’s imagine  that every player has a rating: Beginner,   Intermediate, or Advanced. We want to record the  current rating of each player – and to achieve   that, we simply include in our table  an extra column called Player_Rating.   Notice what’s happening here. Player  jdog21 has a Player_Rating of Intermediate,   but because jdog21 has two rows in the table,  both those rows have to be marked Intermediate.   Player trev73 has a Player_Rating of Advanced,   but because trev73 has four rows in the table, all  four of those rows have to be marked Advanced.   This is not a good design. Why not? Well,  suppose player gila19 loses all her copper coins,   leaving her with nothing in her inventory.  The single entry that she did have in the   Player_Inventory table is now gone. If we try to query the database to find   out what gila19’s Player Rating is, we’re out  of luck. We can no longer access gila19’s Player   Rating because the database no longer knows it.  This problem is known as a deletion anomaly.   And that’s not all. Suppose jdog21 improves  his rating from Intermediate to Advanced.   To capture his new Advanced rating  in the Player_Inventory table,   we run an update on his two records.  But let’s imagine the update goes wrong.   By accident, only one of jdog21’s records gets  updated, and the other record gets left alone.   Now the data looks like this. As far as the database is concerned,   jdog21 is somehow both Intermediate  and Advanced at the same time.   Our table design has left the door open  for this type of logical inconsistency.   This problem is called an update anomaly. Or suppose a new player called tina42 comes along.   She’s a Beginner and she doesn’t have anything  in her inventory yet. We want to record the fact   that she’s a Beginner, but because she  has nothing in her inventory, we can’t   insert a tina42 row into the Player_Inventory  table. So her rating goes unrecorded. This   problem is known as an insertion anomaly. The reason our design is vulnerable to these   problems is that it isn’t in Second Normal  Form. Why not? What is Second Normal Form?   Second Normal Form is about how a table’s non-key  columns relate to the primary key. In our table,   the non-key columns – or to use slightly  different terminology, non-key attributes – are   Item_Quantity and Player_Rating. They are columns  (also called attributes), that don’t belong   to the primary key. As we saw earlier, the primary  key is the combination of Player and Item Type.   Now we’re in a position to give a  definition of Second Normal Form.   The definition we’re going to give is  an informal one which leaves out some   nuances – but for most practical  purposes, that shouldn’t matter.   Informally, what Second Normal Form says  is that each non-key attribute in the table   must be dependent on the entire primary key. How does our table measure up to this definition?   Let’s examine our non-key attributes, which are  the attributes Item_Quantity and Player_Rating.   Does Item_Quantity depend on the entire primary  key? Yes, because an Item_Quantity is about a   specific Item_Type owned by specific  Player. We can express it like this.   The arrow signifies a dependency – or to give  it its proper name, a functional dependency.   This simply means that each value of the thing  on the left side of the arrow is associated with   exactly one value of the thing on the right side  of the arrow. Each combination of Player_ID and   Item_Type is associated with a specific value  of Item_Quantity – for example the combination   of Player_ID jdog21 / Item_Type “amulets”  is associated with an Item_Quantity of 2.   As far as Second Normal Form is  concerned, this dependency is fine,   because it’s a dependency on the entire primary  key. But what about the other dependency?   Does Player_Rating depend on the entire primary  key? No, it doesn’t. Player_Rating is a property   of the Player only. In other words, for any  given Player, there’s one Player_Rating.   This dependency on Player is the problem.  It’s a problem because Player isn’t the   primary key – Player is part of the  primary key, but it’s not the whole key.   That’s why the table isn’t in Second Normal Form,  and that’s why it’s vulnerable to problems.   At what point did our design go wrong, and  how can we fix it? The design went wrong   when we chose to add a Player_Rating column  to a table where it didn’t really belong.   The fact that a Player_Rating is a property  of a Player should have helped us to realise   that a Player is an important concept in its own  right – so surely Player deserves its own table:   Nothing could be simpler than that. A Player  table will contain one row per Player,   and in it we can include as columns the ID of  the player, the rating of the player, as well   as all sorts of other properties of the player  – maybe the player’s date of birth, for example,   maybe the player’s email address. Our other  table, Player_Inventory, can stay as it was.   For both tables, we can say that  there are no part-key dependencies.   In other words, it’s always the case that every  attribute depends on the whole primary key,   not just part of it. And so our  tables are in Second Normal Form.   Now let’s move on to Third Normal Form.  Suppose we decide to enhance the Player table.   We decide to add a new column  called Player_Skill_Level.   Imagine that in this particular multiplayer  game, there’s a nine-point scale for skill level.   At one extreme, a player with skill  level 1 is an absolute beginner;   at the opposite extreme, a player with skill  level 9 is as skilful as it’s possible to be.   And let’s say that we’ve defined exactly how  Player Skill Levels relate to Player Ratings.   “Beginner” means a skill level between  1 and 3. “Intermediate” means a skill   level between 4 and 6. And “Advanced”  means a skill level between 7 and 9.   But now that both the Player_Rating and the  Player_Skill_Level exist in the Player table,   a problem can arise. Let’s say that tomorrow,  player gila19’s skill level increases from 3   to 4. If that happens, we’ll update her row in the  Player table to reflect this new skill level.   By rights, we should also update her Player_Rating  to Intermediate – but suppose something goes   wrong, and we fail to update the Player_Rating.  Now we’ve got a data inconsistency. gila19’s   Player_Rating says she’s a Beginner, but her  Player_Skill_Level implies she’s Intermediate.   How did the design allow this happen? Second  Normal Form didn’t flag up any problems. There’s   no attribute here that depends only partially  on the primary key – as a matter of fact,   the primary key doesn’t have any parts; it’s  just a single attribute. And both Player_Rating   and Player_Skill_Level are dependent on it. But in what way are they dependent on it? Let’s   look more closely. Player_Skill_Level  is dependent on Player_ID.   Player_Rating is dependent on Player ID  too, but only indirectly – like this.   A dependency of this kind is called a transitive  dependency. Player Rating depends on Player Skill   Level which in turn depends on the primary  key: Player ID. The problem is located just   here – because what Third Normal Form forbids is  exactly this type of dependency: the dependency of   a non-key attribute on another non-key attribute. Because Player Rating depends on Player Skill   Level – which is a non-key attribute –  this table is not in Third Normal Form.   There’s a very simple way of repairing the  design to get it into Third Normal Form.   We remove Player Rating from the Player table;  so now the Player table looks like this.   And we introduce a new table  called Player_Skill_Levels.   The Player Skill Levels table tells us everything  we need to know about how to translate a player   skill level into a player rating. Third Normal Form is the culmination of everything   we’ve covered about database normalization so  far. It can be summarised in this way: Every   non-key attribute in a table should depend on  the key, the whole key, and nothing but the key.   If you commit this to memory, and keep it  constantly in mind while you’re designing a   database, then 99% of the time you will  end up with fully normalized tables.   It’s even possible to shorten this guideline  slightly by knocking out the phrase   “non-key” – giving us the revised guideline: every  attribute in a table should depend on the key, the   whole key, and nothing but the key. And this new  guideline represents a slightly stronger flavor of   Third Normal Form known as Boyce-Codd Normal Form.  In practice, the difference between Third Normal   Form and Boyce-Codd Normal Form is extremely  small, and the chances of you ever encountering   a real-life Third Normal Form table that doesn’t  meet Boyce-Codd Normal Form are almost zero.   Any such table would have to have what we call  multiple overlapping candidate keys – which gets   us into realms of obscurity and theoretical  rigor that are a little bit beyond the scope   of this video. So as a practical matter, just  follow the guideline that every attribute in a   table should depend on the key, the whole  key, and nothing but the key, and you can   be confident that the table will be in both  Third Normal Form and Boyce-Codd Normal Form.   In almost all cases, once you’ve normalized  a table this far, you’ve fully normalized   it. There are some instances where this  level of normalization isn’t enough.   These rare instances are dealt with  by Fourth and Fifth Normal Form.   So let’s move on to Fourth Normal Form. We’ll  look at an example of a situation where Third   Normal Form isn’t quite good enough and something  a bit stronger is needed. In our example, there’s   a website called DesignMyBirdhouse.com – the  world’s leading supplier of customized birdhouses.   On DesignMyBirdhouse.com, customers can  choose from different birdhouse models,   and, for the model they’ve selected,  they can choose both a custom color   and a custom style. Each model has its  own range of available colors and styles.   One way of capturing this information  is to put it all the possible   combinations in a single table, like this. This table is in Third Normal Form. The primary   key consists of all three columns: {Model,  Color, Style}. Everything depends on the key,   the whole key, and nothing but the key. And yet this table is still vulnerable   to problems. Let’s look at the rows  for the birdhouse model “Prairie”:   The available colors for the “Prairie”  birdhouse model are brown and beige.   Now suppose DesignMyBirdhouse.com decides  to introduce a third available color for   the “Prairie” model: green. This will mean we’ll  have to add two extra “Prairie” rows to the table:   one for green bungalow, and  one for green schoolhouse.   If by mistake we only add a row for green  bungalow, and fail to add the row for green   schoolhouse, then we have a data inconsistency. Available colors are supposed to be completely   independent of available styles. But our  table is saying that a customer can choose   green only for the bungalow style, not for  the schoolhouse style. That makes no sense.   The prairie birdhouse model is available in green,  so all its styles should be available in green.   Something about the way the table is designed has  allowed us to represent an impossible situation.   To see what’s gone wrong, let’s have a  closer look at the dependencies among Models,   Colors, and styles. Can we say that Color  has a functional dependency on Model?   Actually no, because a specific Model  isn’t associated with just one Color.   And yet it does feel as though Color has some  relationship to Model. How can we express it?   We can say that each Model has a specific set  of available Colors. This kind of dependency is   called a multivalued dependency, and we express  it with a double-headed arrow, like this:   And it’s equally true that each Model  has a specific set of available Styles.   What Fourth Normal Form says is that the only  kinds of multivalued dependency we’re allowed   to have in a table are multivalued dependencies  on the key. Model is not the key; so the table   Model_Colors_And_Styles_Available  is not in Fourth Normal Form.   As always, the fix is to split  things out into multiple tables.   Now, if DesignMyBirdhouse.com expands the range of  Prairie-Model colors to include green, we simply   add a row to the Model_Colors_Available table: And no anomalies are possible.   We’re now ready for Fifth Normal Form, the  last normal form covered in this video.   For our Fifth Normal Form example, we imagine  that there are three different brands of ice   cream available: Frosty’s, Alpine, and Ice  Queen. Each of the three brands of ice cream   offers a different range of flavors: Frosty’s offers vanilla, chocolate,   strawberry, and mint chocolate chip Alpine offers vanilla and rum raisin   Ice Queen offers vanilla,  strawberry, and mint chocolate chip   Now we ask our friend Jason what  types of ice cream he likes.   Jason says: I only like vanilla and chocolate.  And I only like the brands Frosty and Alpine.   We ask our other friend, Suzy, what types of  ice cream she likes. Suzy says: I only like   rum raisin, mint chocolate chip, and strawberry.  And I only like the brands Alpine and Ice Queen.   So, after a little bit of brainwork, we  deduce exactly which ice cream products   Jason and Suzy are willing to eat;  and we express this in a table:   But time passes, tastes change, and at some point  Suzy announces that she now likes Frosty’s brand   ice cream too. So we need to update our table. It won’t come as any surprise that we might get   this update wrong. We might successfully add a  row for Person Suzy – Brand Frosty’s – Flavor   Strawberry, but fail to add a row for Person Suzy  – Brand Frosty’s – Flavor Mint Chocolate Chip.   And this outcome wouldn’t just be wrong – it  would be logically inconsistent – because we’ve   already established that Suzy likes Frosty’s  brand, and likes Mint Chocolate Chip flavor,   and therefore there’s no way she can  dislike Frosty’s Mint Chocolate Chip.   In this example, we went wrong right at the  beginning. At the beginning, we were given   three pieces of information. First, we were told  which brands offered which flavors. Second, we   were told which people liked which brands. Third,  we were told which people liked which flavors.   From those three pieces of information, we  should have simply created three tables.   And that’s all we needed to do. All the  facts of the situation have been represented.   If we ever want to know what  specific products everyone likes,   we can simply ask the database platform,  expressing our question in the form of a   piece of SQL that logically deduces the  answer by joining the tables together.   To sum things up: if we want to ensure  that a table that’s in Fourth Normal   Form is also in Fifth Normal Form, we need  to ask ourselves whether the table can be   logically thought of as being the result  of joining some other tables together.   If it can be thought of that way,  then it’s not in Fifth Normal Form.   If it can’t be thought of that way,  then it is in Fifth Normal Form.   We’ve now covered all the normal forms from First  Normal Form to Fifth Normal Form. Let’s review,   keeping in mind that for a table to comply with  a particular normal form, it must comply with   all the lower normal forms as well. The rules for first normal form are:   1. using row order to convey  information is not permitted   2. mixing data types within the  same column is not permitted   3. having a table without a  primary key is not permitted   4. repeating groups are not permitted The rule for second normal form is:   Each non-key attribute in the table must  be dependent on the entire primary key.   The rule for third normal form is: Each non-key  attribute in a table must depend on the key,   the whole key, and nothing but the key. If we  prefer to drop the phrase “non-key”, we end up   with an even simpler and even stronger version of  third normal form called “Boyce-Codd Normal Form”:   Each attribute in a table must depend on the  key, the whole key, and nothing but the key.   The rule for fourth normal form is that  the only kinds of multivalued dependency   we’re allowed to have in a table are  multivalued dependencies on the key.   Finally, the rule for Fifth Normal Form  is: it must not be possible to describe   the table as being the logical result  of joining some other tables together.   I hope you’ve found this video helpful.  If you have any comments or questions   on what you’ve just seen, by all means  put them in the comments section below.   And if you have any suggestions for other  complex topics that you’d like to see explained   on Decomplexify, again let me know in the  comments. So long, and thanks for watching!