In the question about real vs. fake salt, the answers describe how real salt 'perturbs the encryption algorithm.' I know roughly how initialization vectors work; is this the same concept, or something different entirely?
A salt and an initialization vector are mostly the same thing in the following sense: they are public data, which should be generated anew for each instance (each hashed password, each encrypted message). A salt is about being able to use the same password several times without opening weaknesses; or, if you prefer, preventing an attacker from sharing password attack costs in case the same password could have been used on several instances -- which is all what precomputed (rainbow) tables are about. The point of an IV in, say, symmetric encryption with CBC, is to tolerate the use of the same key to encrypt several distinct messages.
The name "initialization vector" hints at a repetitive process over a given internal state, the IV being what the state is initialized at. For instance, the MD5 hash function is defined as repeated action of a compression function which takes as input the current state (128 bits) and the next message block (512 bits), and outputs the next state value; at the beginning, the state is initialized to a conventional value which is called "the IV". In that sense, most "salts" used in password processing are not "initialization vectors". But this is a bit of an overinterpretation of the expression.
Still, naming things is mostly a matter of Tradition. A "salt" is a kind of IV which:
- is involved in some processing of a password;
- should be distinct for each processing instance (it cannot be a fixed conventional value);
- only needs uniqueness ("it is not repeated"), not uniform selection among the space of possible salts (although uniform random selection is a good and cheap way to get uniqueness with overwhelming probability, assuming that the salts are long enough).
The particulars (how the salt/IV is exactly inserted and at what point in the algorithm) are a red herring.
The answer above is correct when "salt" is discussed in the context of passwords. However, the term "salt" is also used for other uses of random but non-secret values.
For a very rigorous treatment of salt, pertaining to randomness extractors, read Cryptographic Extraction and Key Derivation: The HKDF Scheme. There is some theory there as to why the use of salt is mandatory to obtain generic randomness extractors - although this is not terribly relevant in the case of passwords.