Core CS · DBMS

Numbers do not mean anything by themselves

Q: What is the difference between data and information?

Data is what was recorded; information is data placed in context and processed so that it answers a question. 78, 65, 35, 82, 90 is data. “The DBMS mid-term averaged 70 and 80 per cent of the class passed” is information, and it is the same five numbers. The distinction is about the question being asked, not about the values.

Q: Give me something that is information to one system and data to another.

A delivery address is information to the driver, who uses it to reach a door. The same address is data to the analytics job that groups last month’s orders by pincode. Nothing about the value changed; the question changed. Interviewers ask this to check you can apply the distinction rather than recite “processed data”.

Q: Why is “data is raw, information is processed” an incomplete answer?

Because it leaves out context, which does most of the work. Attaching the column name marks and a student to each number processes nothing at all, and the five values already say far more than they did. Say both halves: context turns values into facts, and processing turns facts into answers.

Q: What is a database?

A shared, integrated collection of related data together with the metadata that describes it, stored so that many users and programs can use it at once without any of them knowing the file layout. Stop at “a collection of related data” and the next question is how that differs from a folder of spreadsheets. The word that saves you is self-describing .

Q: So how is it actually different from a folder of files?

A file hands you bytes and the layout lives in whichever program reads it. A database stores the layout beside the data, enforces constraints on every write, lets several people write at once without corrupting each other, and can undo a half-finished change after a crash. Those four are the answer: self-description, integrity, concurrency, recovery.

Q: What is a DBMS, and what does it do while running one SELECT?

The DBMS is the software layer that owns the files. On one query it parses the statement, checks your privileges, looks up the table and column definitions in the data dictionary, plans how to fetch the rows, reads blocks off disk and cuts them into fields using the offsets the dictionary gave it, then returns the result. Your query never named a file or a byte offset, and that is the point.

Q: What is metadata? Name five kinds.

Data that describes data. Column name, data type, width, whether nulls are allowed, and the constraint the values must satisfy; also the primary key, indexes, owner and grants. Without the type, the byte 4E is equally the number 78 and the letter N. Metadata is what makes a byte a value.

Q: What is the data dictionary, and where is it in a database you have used?

It is the DBMS’s own set of tables holding all the metadata. In PostgreSQL it is pg_catalog , exposed through information_schema . In Oracle it is the dictionary owned by SYS , read through views like USER_TAB_COLUMNS . In SQLite it is the single sqlite_master table. You read all of them with the same SELECT you use on ordinary data.

Q: Who writes to the data dictionary?

The DBMS does, not you. CREATE TABLE and ALTER TABLE are the statements that change it; you read the catalog but you do not hand-edit it. That is why the description can never fall out of step with the data it describes, and it is the difference between metadata inside a database and a comment at the top of a CSV.

Q: Database, DBMS, database system: are they the same thing?

No. The database is the stored data plus its metadata. The DBMS is the software that manages it. The database system is both of those plus the applications and the users on top. Interviewers ask because candidates use “database” for all three, and it stops being a word game the moment you are asked which layer enforces a constraint.

Data is what you recorded. Information is what someone can act on. Everything a DBMS does lives in the gap between them.

Turn five raw marks into an answer →

Data · 78 65 35 82 90

→

Same numbers · mean 70, 80% passed

01 The idea

The gap between a value and a fact

Here are five numbers: 78, 65, 35, 82, 90. You cannot do anything with them. Not because they are wrong or incomplete, but because nothing has said what they count. They could be marks out of 100. They could be ages, or rupees, or the last five readings from a sensor. The numbers do not care and they will not tell you.

Now add one line: these are the marks five students scored in a DBMS mid-term, out of 100, and 40 is the pass mark. Not one digit changed. What changed is that questions became answerable. The class averaged 70. Nikhil topped with 90. Four of the five cleared the pass mark, so 80 per cent of the class passed, and Meera at 35 is the one who did not. The line you added is metadata, and everything after it is arithmetic.

This is also why “a database is a collection of related data” is the answer that gets a follow-up question. A folder of five spreadsheets is a collection of related data. What makes something a database is that the description of the data is stored inside it, next to the data, in a form the software itself can read. Ask a database what its columns are called and it answers. Ask a spreadsheet and you have to go and find the person who made it.

Data is what you recorded. Information is data that has been given context and processed to answer somebody’s question. The bytes on disk are identical in both cases.

DataRecorded values with nothing attached that says what they measure. 35 is data. So is Meera. Data is whatever survives if you delete every column heading.

InformationData that has been given context and processed so that it answers a question somebody actually asked. “Meera is the only student below the pass mark” is information.

MetadataData that describes data: a column’s name, its type, its width, whether it may be empty, and the rule its values must obey. It is what turns a byte into the number 78.

02 Worked example

One question, and the five layers under it

You type one line: SELECT student, marks FROM marks WHERE marks < 40;. Nothing in it names a file, a disk block or a byte offset. Follow the same five values from section 01 on their way to becoming an answer.

stored7865358290data · five integers, nothing attached

answer7865358290information · the only mark below the pass mark of 40

Those two rows hold the same bytes. Everything that happened between them happened in five layers, and only the first and the last are visible to you.

Your queryone line of SQL: a table and a condition, nothing else

→

DBMSparses it, checks you are allowed to read MARKS, plans the fetch

→

Data dictionarythe DBMS’s own tables: marks is a 2-byte integer, second in the row

→

Stored blocks8 KB of bytes on disk, none of them labelled

→

ResultMeera, 35 — one row you can act on

The highlighted node is the one students skip. Without the data dictionary the storage engine has a block of bytes and no way to cut it into fields, and the DBMS has a table name it cannot resolve. Metadata is not documentation sitting beside the database. It is a working part of every single read.

Take the middle three nodes away and you are back to a file plus a person who has to know its layout. That is how it was done before databases: the byte offsets lived inside every program that read the file, so the storage and the programs could never change independently. Moving that knowledge into the dictionary is what breaks the coupling, and it has a name. Physical data independence is the ability to change how data is stored without rewriting the code that reads it.

03 Mechanics

What the layer buys you, row by row

The interview version of this section is “file system versus DBMS”. Answer it as a list of properties, not a list of adjectives. Each row below is one thing that goes wrong with plain files and the named mechanism that fixes it.

What you are relying on	A folder of files	A DBMS
Where the layout is written down	`inside every program that reads it`	`once, in the data dictionary`
Two people writing at the same time	`last writer wins`	`transactions and locking`
A crash halfway through a change	`a half-written file`	`atomicity: all of it or none of it`
Stopping a bad value getting in	`whatever the writer remembered to check`	`types, NOT NULL, CHECK, foreign keys`
Asking a question nobody planned for	`write a program`	`write a query`
Finding one row among a million	`read the whole file`	`an index the planner chooses`
Who may see which part	`the whole file, per operating-system user`	`GRANT, per table and per role`

The first row is the one everything else rests on, so it is worth opening up. Here is what the dictionary records about a single column, marks.marks, and what stops working if that entry is missing.

What is recorded	For marks.marks	What breaks without it
Column name	`marks`	You address the value by byte offset instead, in every program that reads it.
Type and width	`SMALLINT, 2 bytes`	The byte 4E is the number 78 and also the letter N. Nothing decides between them.
Position in the row	`second`	The storage engine has a block of bytes and no way to cut it into fields.
Nullability	`NOT NULL`	“Has not sat the exam yet” and “scored zero” become the same row.
Constraint	`CHECK (marks BETWEEN 0 AND 100)`	A typo of 350 is stored, and every average computed after it is wrong.
Privileges	`teacher may UPDATE, student may SELECT`	Anyone who can open the file can edit any mark in it.

Every one of those entries is written by the DBMS itself, when you run CREATE TABLE or ALTER TABLE, and read back with an ordinary SELECT. That is why the description can never drift away from the data it describes. A comment at the top of a spreadsheet can, and usually has.

05 Cheat sheet

Say this, not that

Every term here has a short answer that is correct and a shorter one that is correct but hollow. The hollow one is not wrong, which is exactly why it invites the follow-up question in the third column.

Term	Say this	The answer that invites a follow-up
Data	`recorded values, no question attached`	`“numbers”` — then a name or a date is not data, which you do not want to defend.
Information	`data given context and processed to answer a question`	`“processed data”` — true and empty. The next question is what processing, and context does half the work with none.
Database	`an integrated, shared, self-describing store of data and its metadata`	`“a collection of related data”` — so is a folder of spreadsheets, and that is the follow-up.
DBMS	`the software that owns the files and answers queries about them`	`“software to manage a database”` — circular. Name what it does: dictionary, constraints, transactions, recovery, access control.
Metadata	`names, types, widths, nullability, constraints, owners, indexes`	`“data about data”` — the follow-up is “such as?” and the list above is the whole answer.
Data dictionary	`the DBMS’s own tables holding all of that metadata`	`“where the schema is kept”` — vague. Name a real one you have queried.
Database system	`the database, plus the DBMS, plus the applications and users on top`	`“same as a database”` — then you cannot say which layer enforces a constraint.

The self-describing testOpen a CSV and somebody has to tell you what column three is. Open a database and you can ask it, because the description is stored in the same system as the data, in tables you query the same way. That single property is most of the difference between a file and a database.

The labels swapA delivery address is information to the driver and data to the job that counts orders per pincode. Nothing about the value changed; the question did. Any definition that fixes a value permanently as one or the other is wrong.

The layer is not freeDictionary lookups, privilege checks and transaction bookkeeping are work that a plain file read does not do. For one program reading one flat file start to finish, raw file I/O wins. You pay the overhead for the guarantees, not for speed.

06 Where & why

Where the metadata actually lives

The data dictionary is not an abstraction from a textbook. It is a set of tables with names, in every database you have ever opened, and you can read them today. An answer that names one lands very differently from one that says “the metadata is stored somewhere in the system”.

PostgreSQL · MySQL

The dictionary is queried with ordinary SQL

PostgreSQL keeps its catalogs in the pg_catalog schema and exposes the standard view of them as information_schema. MySQL exposes information_schema too, and SHOW COLUMNS FROM marks is the shortcut. Same SELECT, same result grid, same permissions model as the data itself.

Oracle

Where the phrase comes from

Oracle calls it the data dictionary outright: base tables owned by SYS, read through views such as USER_TAB_COLUMNS and ALL_TABLES. The USER_, ALL_ and DBA_ prefixes are themselves an access-control decision expressed in metadata.

SQLite

One file, and it still describes itself

sqlite_master holds one row per table, index, view and trigger, and its sql column stores the original CREATE statement text. The whole database, data and dictionary together, is a single file on disk. Self-describing does not require a server.

Excel · CSV

The case where the metadata lives in someone’s head

A column headed marks holding the value 35 has a heading and nothing else: no type, no constraint, no owner. The rule that marks run 0 to 100 exists only as the habit of whoever maintains the sheet, so nothing stops a typo of 350 and nothing records who typed it. Every argument for a DBMS starts here.

If you can name where the metadata lives in one database you have actually used, and say what you found there, the “what is a DBMS” question is over in a sentence. If you cannot, every answer you give is a definition somebody else wrote.

07 Interview questions

What they actually ask

This is the first question in most DBMS interviews and the one candidates most often answer from memory. The give-away is a definition that cannot survive one follow-up. Answer each of these in about twenty seconds, out loud.

What is the difference between data and information?

Data is what was recorded; information is data placed in context and processed so that it answers a question. 78, 65, 35, 82, 90 is data. “The DBMS mid-term averaged 70 and 80 per cent of the class passed” is information, and it is the same five numbers. The distinction is about the question being asked, not about the values.

Give me something that is information to one system and data to another.

A delivery address is information to the driver, who uses it to reach a door. The same address is data to the analytics job that groups last month’s orders by pincode. Nothing about the value changed; the question changed. Interviewers ask this to check you can apply the distinction rather than recite “processed data”.

Why is “data is raw, information is processed” an incomplete answer?

Because it leaves out context, which does most of the work. Attaching the column name marks and a student to each number processes nothing at all, and the five values already say far more than they did. Say both halves: context turns values into facts, and processing turns facts into answers.

What is a database?

A shared, integrated collection of related data together with the metadata that describes it, stored so that many users and programs can use it at once without any of them knowing the file layout. Stop at “a collection of related data” and the next question is how that differs from a folder of spreadsheets. The word that saves you is self-describing.

So how is it actually different from a folder of files?

A file hands you bytes and the layout lives in whichever program reads it. A database stores the layout beside the data, enforces constraints on every write, lets several people write at once without corrupting each other, and can undo a half-finished change after a crash. Those four are the answer: self-description, integrity, concurrency, recovery.

What is a DBMS, and what does it do while running one SELECT?

The DBMS is the software layer that owns the files. On one query it parses the statement, checks your privileges, looks up the table and column definitions in the data dictionary, plans how to fetch the rows, reads blocks off disk and cuts them into fields using the offsets the dictionary gave it, then returns the result. Your query never named a file or a byte offset, and that is the point.

What is metadata? Name five kinds.

Data that describes data. Column name, data type, width, whether nulls are allowed, and the constraint the values must satisfy; also the primary key, indexes, owner and grants. Without the type, the byte 4E is equally the number 78 and the letter N. Metadata is what makes a byte a value.

What is the data dictionary, and where is it in a database you have used?

It is the DBMS’s own set of tables holding all the metadata. In PostgreSQL it is pg_catalog, exposed through information_schema. In Oracle it is the dictionary owned by SYS, read through views like USER_TAB_COLUMNS. In SQLite it is the single sqlite_master table. You read all of them with the same SELECT you use on ordinary data.

Who writes to the data dictionary?

The DBMS does, not you. CREATE TABLE and ALTER TABLE are the statements that change it; you read the catalog but you do not hand-edit it. That is why the description can never fall out of step with the data it describes, and it is the difference between metadata inside a database and a comment at the top of a CSV.

Database, DBMS, database system: are they the same thing?

Hard

A legacy system stores each order as a fixed-width binary record: bytes 0 to 3 the order id, 4 to 33 the customer name, 34 to 37 the amount in paise, byte 38 a status code. Six programs read this file. Describe everything that must change to widen the customer name from 30 bytes to 50, then describe what changes in a database holding the same orders in a table, and name the property that accounts for the difference.

Follow-up

The database answer is not “nothing changes”. One thing does, and being honest about which one is what separates a real answer from a slogan.

Show the hint

Find every place the number 30 is written down in each design before you write a word about the migration.