We consider the problem of characterizing the per-node throughput
scaling in arbitrary extended wireless networks.  Recently, Ozgur,
Leveque, and Tse (2007) obtained a complete characterization of the
throughput scaling in random networks (i.e., nodes are placed in a
square region uniformly at random) under a fast-fading channel
model. They proposed a hierarchical cooperative communication scheme
to establish the achievability. Their results and proof techniques
are, however, strongly dependent on the "regularity" induced with high
probability by the random node placement. We propose a more general
and a different hierarchical cooperative communication scheme that
works for arbitrarily placed nodes with a minimum-separation
requirement. We show that the proposed scheme achieves exactly the
same per-node throughput scaling as in Ozgur et al., showing that much
less regularity is necessary for successful hierarchical cooperation.
Our results hold under both fast- and slow-fading channel model.  For
small power path-loss exponents $\alpha \in (2,3]$, we show that the
scheme is order optimal for all node placements with minimum-separation 
requirement. We also propose a cooperative multi-hop scheme that 
"interpolates" between hierarchical cooperation and multi-hopping 
depending on the level of "regularity" in the node placement. We 
establish that for certain node placements the scheme is order
optimal, and strictly better than multi-hop, for every $\alpha >2$.