A comprehensive survey of Distributed Data Management algorithms—from structured trees to gossip protocols—and our proposed Adaptive Gossip-Hierarchy (AGH) algorithm.
A tree-based aggregation service that treats the sensor network as a distributed database, answering SQL-like queries using slot-synchronized collection phases.
Based on mass conservation — each node splits its (sum, weight) pair and gossips half to a random neighbor. The ratio s/w at every node converges exponentially to the true global average.
Cluster Heads compute J(A,B) = |A∩B| / |A∪B|. If similarity exceeds threshold δ (e.g., 0.85), redundant packets are suppressed and a single representative set is forwarded.
Cluster Heads apply the F-statistic to compare inter-group variance against intra-group noise. If H₀ is accepted (no significant difference), all readings are replaced by a single (μ, σ²) summary.
A self-healing hybrid that operates in TAG mode for optimal efficiency, then seamlessly transitions to gossip-based consensus during network failures and re-integrates automatically.
Qualitative evaluation across critical deployment dimensions.
| Algorithm | Network Overhead | Data Accuracy | Fault Tolerance | Best For |
|---|---|---|---|---|
| TAG (Structured) | O(n) per epoch | ✅ Exact | ⚠️ Low | Static, energy-critical deployments |
| Push-Sum (Gossip) | O(n log n) | ⚠️ Approximate | ✅ Highest | Mobile networks, high churn |
| Jaccard (Statistical) | Low (filtered) | ✅ High | ⚠️ Moderate | Categorical / multiset redundancy |
| ANOVA (Statistical) | Moderate | ✅ Stat-significant | ⚠️ Moderate | Dense sensor grids with redundancy |
| AGH Our Proposal | Adaptive | ✅ High (hybrid) | ✅ 95% @ 20% failures | Mission-critical IoT, dynamic nets |
Different algorithms govern different architectural tiers of a production data solution.