Database Design – Difference Between Field Specification and Field Domain

I personally would use a model similar to the following:

The product table would be pretty basic, your main product details:

create table product
(
  part_number int, (PK)
  name varchar(10),
  price int
);
insert into product values
(1, 'product1', 50),
(2, 'product2', 95.99);

Second the attribute table to store the each of the different attributes.

create table attribute
(
  attributeid int, (PK)
  attribute_name varchar(10),
  attribute_value varchar(50)
);
insert into attribute values
(1, 'color', 'red'),
(2, 'color', 'blue'),
(3, 'material', 'chrome'),
(4, 'material', 'plastic'),
(5, 'color', 'yellow'),
(6, 'size', 'x-large');

Finally create the product_attribute table as the JOIN table between each product and its attributes associated with it.

create table product_attribute
(
  part_number int, (FK)
  attributeid int  (FK) 
);
insert into product_attribute values
(1,  1),
(1,  3),
(2,  6),
(2,  2),
(2,  6);

Depending on how you want to use the data you are looking at two joins:

select *
from product p
left join product_attribute t
  on p.part_number = t.part_number
left join attribute a
  on t.attributeid = a.attributeid;

See SQL Fiddle with Demo. This returns data in the format:

PART_NUMBER | NAME       | PRICE | ATTRIBUTEID | ATTRIBUTE_NAME | ATTRIBUTE_VALUE
___________________________________________________________________________
1           | product1   | 50    | 1           | color          | red
1           | product1   | 50    | 3           | material       | chrome
2           | product2   | 96    | 6           | size           | x-large
2           | product2   | 96    | 2           | color          | blue
2           | product2   | 96    | 6           | size           | x-large

But if you want to return the data in a PIVOT format where you have one row with all of the attributes as columns, you can use CASE statements with an aggregate:

SELECT p.part_number,
  p.name,
  p.price,
  MAX(IF(a.ATTRIBUTE_NAME = 'color', a.ATTRIBUTE_VALUE, null)) as color,
  MAX(IF(a.ATTRIBUTE_NAME = 'material', a.ATTRIBUTE_VALUE, null)) as material,
  MAX(IF(a.ATTRIBUTE_NAME = 'size', a.ATTRIBUTE_VALUE, null)) as size
from product p
left join product_attribute t
  on p.part_number = t.part_number
left join attribute a
  on t.attributeid = a.attributeid
group by p.part_number, p.name, p.price;

See SQL Fiddle with Demo. Data is returned in the format:

PART_NUMBER | NAME       | PRICE | COLOR | MATERIAL | SIZE
_________________________________________________________________
1           | product1   | 50    | red   | chrome   | null
2           | product2   | 96    | blue  | null     | x-large

As you case see the data might be in a better format for you, but if you have an unknown number of attributes, it will easily become untenable due to hard-coding attribute names, so in MySQL you can use prepared statements to create dynamic pivots. Your code would be as follows (See SQL Fiddle With Demo):

SET @sql = NULL;
SELECT
  GROUP_CONCAT(DISTINCT
    CONCAT(
      'MAX(IF(a.attribute_name = ''',
      attribute_name,
      ''', a.attribute_value, NULL)) AS ',
      attribute_name
    )
  ) INTO @sql
FROM attribute;

SET @sql = CONCAT('SELECT p.part_number
                    , p.name
                    , ', @sql, ' 
                   from product p
                   left join product_attribute t
                     on p.part_number = t.part_number
                   left join attribute a
                     on t.attributeid = a.attributeid
                   GROUP BY p.part_number
                    , p.name');

PREPARE stmt FROM @sql;
EXECUTE stmt;
DEALLOCATE PREPARE stmt;

This generates the same result as the second version with no need to hard-code anything. While there are many ways to model this I think this database design is the most flexible.

What you're thinking about is either "row modelling" or "entity-attribute-value" (EAV) - See Wikipedia - either of which is a common subject for religious wars. I won't get too far into the pros vs. cons. If you read up on EAV and row modeling you'll see lots of discussion on that.

What I will say is that you should model all of the common predicates using properly normalized tables and rely on EAV (or row modelling) only for the customizable columns (if at all). This is how many highly customizable packaged solutions deal with allowing users to add their own attributes to existing objects. It isn't clean, easy or efficient, but it is probably the best compromise for allowing this particular type of customization when you have no way of reasonably predicting what the customized columns might be at design time.