DynaMIC: Dynamic Context Monitoring Framework

• DynaMIC is a dynamic context monitoring framework designed to optimize context caching in IoT by leveraging a two-stage pipeline for evaluation and cache management.
• It integrates a Context Evaluation Engine using Multi-Attribute Utility and Probability of Access with a Context Management Module employing DST-based fusion to ensure optimal caching decisions.
• Performance evaluations on smart city datasets demonstrate significant improvements in cache hit ratios, reduced expiry rates, and lower response times compared to state-of-the-art methods.

The Dynamic Context Monitoring Framework (DCMF), also known as “DynaMIC”, is a comprehensive context caching and management pipeline designed to address the volatile nature of context data in context-aware Internet of Things (IoT) environments. The framework enhances context freshness and cache hit ratios by employing a two-stage architecture that integrates dynamic context evaluation and hybrid evidence-based cache management. DynaMIC has been evaluated within the Context-as-a-Service (CoaaS) paradigm on real-world smart city traffic and roadwork datasets, demonstrating significant gains in cache efficacy and timeliness over state-of-the-art competitors (Manchanda et al., 25 Apr 2025).

1. Architectural Structure

DynaMIC operates as a subsystem within a Context Management Platform (CMP) such as CoaaS or FIWARE Orion. Its architecture is split into two tightly integrated components:

• Context Evaluation Engine (CEE): Interfaces with the CMP’s Context Query Engine to receive context queries from context consumers (CCs), maintains historical and real-time logs of context item (CI) accesses, extracts metadata (Quality of Service (QoS), Quality of Context (QoC), Cost of Context (CoC), timeliness, Service Level Agreements (SLAs)), and computes both a MAUT-based utility and Probability of Access (PoA) for each CI.
• Context Management Module (CMM): Maintains an in-memory cache of context items and optionally their associated context attributes, monitors PoA and context freshness (CF) metrics for all cached CIs, and utilizes a hybrid Dempster–Shafer Theory (DST) mechanism to determine optimal cache management decisions: retain, refresh, or evict.

A high-level depiction of system interactions is as follows:

CEE	Priority List →	CMM
CQ Logs ⇵		Cache

This split enables decoupling of context prioritization from cache decision policies, making the system adaptable to fluctuations in access patterns and context volatility (Manchanda et al., 25 Apr 2025).

2. Evaluation Process and Mathematical Foundations

The core of DynaMIC’s adaptability lies in its quantitative evaluation of both access likelihood and freshness for each context item.

A. Multi-Attribute Utility (MAUT):

For each CI, the engine computes a utility score as follows:

$$U(CI_i) = \sum_{j=1}^{m} w_j u_j(CI_i) \qquad \sum_{j=1}^m w_j = 1$$

where $u_j(CI_i)$ are normalized scores for individual attributes (QoS, QoC, CoC, timeliness, SLA-compliance) and $w_j$ are their attribute weights—typically determined using the Analytic Hierarchy Process (AHP).

B. Probability of Access (PoA):

PoA models CI demand as a weighted blend of historical hits and recent queries:

$$\text{PoA}(CI_i) = \alpha \frac{h(CI_i)}{\sum_{k=1}^{n} h(CI_k)} + (1-\alpha) \frac{q(CI_i)}{N}$$

$\alpha$ modulates the weight given to historical ($h(CI_i)$, Poisson-modeled) vs. real-time ($q(CI_i)$, Gaussian-modeled within window $T$) query rates.

C. Hybrid Prioritization:

CIs are ranked by a composite priority:

$\beta U(CI_i) + (1-\beta)\text{PoA}(CI_i)$

delivering an ordered priority list forwarded to the CMM (Manchanda et al., 25 Apr 2025).

3. Hybrid Evidence-Based Cache Management

Cache retention decisions in DynaMIC are implemented using a DST-based fusion of context freshness and access estimation. For each CI:

• Freshness Metric:

$\text{CF}(CI_i) = e^{-\lambda \Delta t_i}$

with $u_j(CI_i)$0 the elapsed time since last refresh and $u_j(CI_i)$1 a decay constant.

• Basic Belief Assignments:

DST combines PoA and CF as independent sources (“evidence”), assigning to each CI beliefs for “Cache” and “Evict” actions:

$u_j(CI_i)$2

• Conflict and Combination:

The DST conflict coefficient,

$u_j(CI_i)$3

and the combined belief in action $u_j(CI_i)$4:

$u_j(CI_i)$5

• Decision Procedure:

$u_j(CI_i)$6

with $u_j(CI_i)$7 and $u_j(CI_i)$8 typically parameterized from the empirical mean and variance of CF for dynamic thresholds. This hybrid inference allows DynaMIC to robustly arbitrate between access demand and data staleness (Manchanda et al., 25 Apr 2025).

4. Implementation Details and Workflows

DynaMIC was instantiated on a CoaaS platform via microservices in Java and Python, accessed via REST and gRPC interfaces. The experimental setup included:

• Data Sources: Nerian 3D depth-camera captures (mixed real and synthetic), public traffic and roadworks datasets with 268,600 data points.
• Workload Simulation: Context Query Simulator (Gaussian arrivals), 70,000 CQ events over 24h, spanning low/medium/high loads (Poisson $u_j(CI_i)$9 = 0.5–2 q/s), 30 context services, ~10,000 simulated CCs.
• Cache Operations: Two main workflow procedures govern CEE and CMM, as summarized in the following pseudocode blocks:

$w_j$4 All decision parameters are subject to runtime adaptation to track context volatility (Manchanda et al., 25 Apr 2025).

5. Comparative Performance Evaluation

DynaMIC’s performance was benchmarked against context-aware cache management techniques including m-CAC, m-Greedy, and m-Myopic. Key outcomes included:

Scenario (Load/Condition)	Cache Hit Ratio (CHR)	Cache Expiry Ratio (CER)	Avg. Response Time $w_j$0	Throughput (req/min)
High CF	80% (DCMF) vs. 65%	5% vs. 18–25%	120 ms vs. 190 ms	—
High PoA (100 q/s)	85% vs. 58–68%	—	150 ms vs. 250–400 ms	—
Balanced	82% vs. 68%	6% vs. 15%	120 ms vs. 190 ms	—
Scalability (low/med/high)	—	—	—	53.8/50.5/47.8 (DCMF); 45.8/43.1/41.8 (m-CAC)

• In high-freshness scenarios, DynaMIC improved CHR by 23.1%, reduced CER by up to 60%, and offered 36.8% lower average latency relative to m-CAC.
• Under high access probability, DynaMIC delivered up to 85% hit ratio and response times as low as 14 ms (at 1000 MB cache) (Manchanda et al., 25 Apr 2025).

These results underscore the advantage of joint MAUT-DST context modeling over pure frequency- or freshness-based heuristics.

6. Limitations and Prospective Enhancements

The DST-based fusion and dynamic evaluation in DynaMIC introduces marginally increased memory and CPU overhead relative to simpler caching schemes. Cache refresh frequency and the cost of DST computation necessitate deliberate parameter tuning: utility weights $w_j$1, blending factors $w_j$2, decay $w_j$3, and decision thresholds. Overly aggressive parameterization can result in unnecessary refreshes or stale cache retention.

Anticipated extensions include:

• Integration of machine learning approaches (e.g., reinforcement learning, LSTM forecasting) for predictive PoA and CF estimation.
• Online adaptive thresholding to automatically self-tune fusion and eviction constraints.
• Deployment of energy-conscious variants suitable for resource-constrained edge environments.
• Enhanced privacy/security modules for sensitive contexts.

These directions reflect the broader trend toward self-adaptive, heterogeneity-aware context caching in IoT platforms (Manchanda et al., 25 Apr 2025).

7. Significance and Application Scope

DynaMIC demonstrates a two-stage CEE→CMM pipeline that fuses long-term and immediate CQ demand with stringent freshness management via DST, validated on diverse, high-churn smart city scenarios. Its empirical results—20–25 percentage-point gains in cache hit ratio, reductions of expired items up to 60%, and 30–40% lower response times—establish DynaMIC as a scalable and effective solution for real-time, context-aware IoT platforms requiring robust, adaptive caching and freshness assurance (Manchanda et al., 25 Apr 2025).