Well, you could certainly just assume that all friendships are symmetric and store that friendship only once, but that would mean that when you want to query for all of Taher's friends, you have to look for his ID in either column.
Alternately you could have a separate table of relationship ID's, and then a one-to-many table of relationship to user. This has the advantage that it would allow multi-person relationships if someday you want that, and would let you add meta-data about the relationship (when it started, who suggested it, whatever).
User
Id Name
1 Taher
2 Deepak
Relationship
Id StartDate
1 2010-08-23
UserRelationship
RelationshipId UserId
1 1
1 2
On the other hand, on Facebook, for example, I can "friend" someone and they can decide not to friend me back. If you don't have the "redundant" approach that you're using now, how will you represent that not-yet-reciprocal friendship?