A variable in PDI is a piece of user-supplied information that you can use dynamically and programmatically in a variety of different scopes. A variable can be local to a single step, or be available to the entire JVM (Java Virtual Machine) that PDI is running in.
PDI variables can be used in both Basic concepts of PDI transformation steps and job entries. You define variables with the Set Variable step and Set Session Variables step in a transformation, by hand through the kettle.properties file, or through the Set Environment Variables dialog box in the Edit menu.
The Get Variable and Get
Session Variables steps can explicitly retrieve a value from a variable, or you
can use it in any PDI
text field which has the dollar sign 
icon next to it by using a metadata string in either the Unix or Windows formats:
- ${VARIABLE}
- %%VARIABLE%%
Both formats can be used and even mixed. In fact, you can create variable recursion by alternating between the Unix and Windows syntax. For example, if you wanted to resolve a variable that depends on another variable, then you could use this example: ${%%inner_var%%}.
You can also use ASCII or hexadecimal character codes in place of variables, using the same format: $[hex value]. This makes it possible to escape the variable syntax in instances where you need to put variable-like text into a variable. For instance if you wanted to use ${foobar} in your data stream, then you can escape it like this: $[24]{foobar}. PDI will replace $[24] with a $ without resolving it as a variable.