Deployment at Finite Scale

From Theory to 6G Prototypes

The theoretical chapters set the foundation; the subpacketization breakthroughs of §14.2-14.3 make deployment feasible. This section addresses how coded caching will actually roll out in 6G systems — prototypes, standardization, operational concerns.

⚠️Engineering Note

Coded Caching in 5G / 6G Roadmaps

Current standardization status (as of 2026):

5G NR (Rel-15-17). Basic multicast primitives (MBMS). Cache-aware extensions exist as study items but not in standard.
5G-Advanced (Rel-18-19). Enhanced multicast; user grouping for broadcast. Supports cache-aware content partitioning.
6G vision (Rel-20+). Distributed coded-caching delivery in fog/cell-free massive MIMO. Cache-aware MBSFN extensions with coding primitives.
ORAN. Open RAN architecture supports pluggable caching layers. Operators can add coded caching without vendor lock-in.

Deployment timeline (estimated):

2024-2025: Vendor prototypes at small scale ( $K = 10\text{-}50$ ).
2026-2028: Testbeds, partial deployment in stadiums / venues ( $K = 100$ ).
2029+: 6G standardization; production rollout.

The CommIT group has published several papers on deployable coded caching and contributes to 3GPP standardization (informally).

Practical Constraints

•
3GPP 5G Rel-17: multicast support, cache-aware extensions studied
•
Rel-18-19 (5G-Advanced): enhanced multicast primitives
•
6G Rel-20+: cache-aided delivery standardization planned
•
ORAN: modular caching layers supported

Definition:
Practical Coded-Caching Scheme Requirements

A practical coded-caching scheme (for 6G deployment) must satisfy:

Polynomial subpacketization. $F = O(K^c)$ for small constant $c \leq 3$ .
Near-optimal rate. Rate within factor 2 of MAN.
Robust to user dynamics. Users arriving/departing within a session shouldn't break the scheme.
Privacy-preserving. Demands not leaked to other users (Chapter 12 techniques).
Heterogeneous-user-friendly. Adapts to different cache sizes (Chapter 13).
Standards-compatible. Uses 3GPP multicast primitives, not exotic protocols.

Not all known PDA schemes satisfy all requirements. Engineering practice: pick a subset that meets your deployment target.

Example: Designing a 6G Coded-Caching Service

A 6G operator wants to deploy cache-aided video delivery for 20,000 users per cell. Library: 50,000 movies. Per-user cache: 100 movies. Target: 4K video streaming at <100 ms latency. Choose scheme parameters.

Solution

Network scaling

$K = 20{,}000$ is too large for single-cluster MAN. Partition into clusters of $K_g = 100$ (200 clusters). Spatial reuse across clusters.

Within-cluster scheme

MAN with $K_g = 100$ , $\mu = 100/50{,}000 = 0.002$ . $t = 0.2$ . Too small — not useful. Try $\mu = 0.01$ (500 movies in cache): $t = 1$ . Need $\binom{100} {1} = 100$ subfiles per file. Feasible. Polynomial PDA: $F \approx 100^2 = 10^4$ . Each 100MB movie becomes 10KB subfiles. Acceptable.

Delivery rate

With $t = 1$ , $K_g = 100$ : per-cluster rate = $100 \cdot 0.99/2 \approx 49.5$ files. Per-user rate $= 49.5/100 = 0.495$ files per round. With 4K video at 25 Mbps and round = 1 sec: per user $\approx 50 \cdot 25 = 1250$ Mbps served — plenty.

Latency

Placement: off-peak (days). Delivery round: $\sim$ 100ms target; 50 transmissions fit in this window on a 5G NR frame. Meets latency target.

Conclusion

PDA-based scheme with $K_g = 100$ , $M = 500$ , polynomial $F = 10^4$ satisfies all 6G requirements. Deployable today with ORAN-compatible modifications.

Hybrid Architectures in Practice

Real 6G deployments will use hybrid architectures combining:

User-side cache (phone / device): $M_u \sim 10$ GB.
Edge cache (RRU / AP): $M_e \sim 1$ TB per AP.
Fog cache (DU): $M_f \sim 10$ TB per DU.
Cloud cache (CU): effectively unlimited.

Different content tiers at different caches:

Hottest content: user-side (instant access).
Warm content: edge (sub-ms access via D2D or short-link).
Cool content: fog (low-latency fetch).
Cold content: cloud (long-latency, bandwidth-limited).

Coded caching operates differently at each tier. The CommIT framework provides a unified view: each tier has a caching gain $t_{\text{tier}} = K_{\text{tier}} M_{\text{tier}}/N_{\text{tier}}$ ; aggregate rate is a product of per-tier gains. Design becomes a hierarchical optimization.

Common Mistake: Don't Price Based on Theoretical Rate Alone

Mistake:

Quoting a coded-caching system at the theoretical MAN rate without accounting for scheme / deployment overhead.

Correction:

Deployed systems typically achieve 50-70% of theoretical MAN rate due to:

Polynomial vs exponential $F$ : $<$ 2× rate gap.
Cluster boundaries: non-zero overhead.
Scheduling: not perfectly parallel.
Imperfect placement: user arrivals disrupt structure.

Engineering reality: budget 50-70% of theoretical rate; use that for capacity planning. The difference is the "deployment tax" and is unavoidable in practice.

Quick Check

Why can't the MAN scheme be deployed directly for $K = 1000$ ?

It's mathematically incorrect at large K.

Subpacketization $\binom{K}{t}$ becomes astronomical.

The rate formula is unstable at large K.

XOR decoding fails at large K.

Correction:

Subpacketization

\binom{K}{t}

becomes astronomical.

Correct. At $K = 1000$ , $\mu = 0.1$ : $\binom{1000}{100} \approx 10^{139}$ subfiles per file. Each subfile would be sub-bit size for typical files. Impossible to implement.

Historical Note: Evolution of Subpacketization Solutions

2014–2025

The subpacketization problem was recognized early but took years to solve:

2014: MAN scheme introduced. Exponential $F = \binom{K}{t}$ noted but downplayed.
2016: Shanmugam-Ji-Tulino-Llorca-Dimakis-Caire — first serious finite-length analysis. Acknowledges the problem.
2017: Yan-Cheng-Tang-Chen — PDA framework; first polynomial- $F$ constructions.
2018-2020: Explosion of PDA variants; CommIT group's graph-coloring schemes; Salehi-Tölli-Shariatpanahi-Peiker practical implementations.
2021-2024: Prototype implementations at $K = 50\text{-}100$ in research testbeds.
2025+: 6G standardization of cache-aware delivery.

The progression from "theoretical impossibility" to "deployable technology" took ~10 years. The CommIT group has been central throughout — the subpacketization problem was arguably their main practical research contribution to coded caching.

Graph-Coloring Delivery (CommIT)Chapter Summary