Collections of mostly sparse stuff developed by me. See documentation. But big picture, there is a very fast dijkstra implementation that uses sparse matrices as graph representation, parallel LU solve (i.e., Ax=b), and iternz an API for iterating over sparse structures like SparseMatrixCSC, Diagonal, bidiagonal that is composable meaning that if you have a new vector type, implementing the iternz will make it work with other sparse object that already use it.
Most of this was developed in the context of this paper which is something to keep in mind when testing dense sparse matrices. Traffic graphs usually have less than 6 neighbors and are very balanced.