Chapter Summary

Chapter Summary

Key Points

• 1.

  The D2D caching network has $n$ users acting as both transmitters and receivers, each with a cache of $M$ files from a library of $N$ files. Delivery occurs peer-to-peer via short-range D2D links — no central server.

• 2.

  Ji-Caire-Molisch scaling law (CommIT 2016): per-user throughput scales as $\Theta(\mu)$, independent of $n$. Caching effectively converts the local D2D network into a globally distributed library; the supply scales with demand.

• 3.

  D2D dominates infrastructure at scale. Infrastructure per-user throughput: $O(\log n/n) \to 0$. D2D per-user throughput: $\Theta(M/N)$, constant. Crossover around $n \sim 100$; for dense networks, D2D is strictly better.

• 4.

  Model assumptions are critical. The scaling law holds for random geometric graphs, protocol interference, random uniform demands, fixed memory ratio. Different models give different scalings; don't overclaim.

• 5.

  Placement strategies. Random uniform (simple, asymptotically optimal); popularity-proportional (better for Zipf); combinatorial (MAN-style, enables coded multicasting in Ch 11).

• 6.

  Coverage vs diversity tradeoff. Placement must balance per-file coverage (many users cache the same file) against neighborhood diversity (different users cache different files). Randomized placement with slight correlation hits the right balance asymptotically.

• 7.

  Engineering barriers. D2D is standardized (ProSe, Sidelink) but not widely deployed due to operator economics, user incentives, battery cost, and privacy concerns. Chapter 12 addresses the privacy angle.