Edge Computing For IoT Mini Review
Uploaded by
Afzaal juttEdge Computing For IoT Mini Review
Uploaded by
Afzaal juttEdge Computing for IoT: Reducing Latency and Cloud Cost
Prepared by ChatGPT September 14, 2025
______________________________________________________________________
ABSTRACT
This mini review surveys how edge computing shifts computation closer to sensors to reduce
latency, bandwidth, and cost. We summarize architectural patterns (cloud only, edge cloud
split, and fully local), discuss security and reliability trade offs, and outline a lightweight
evaluation approach for teams comparing options for time sensitive Internet of Things (IoT)
workloads.
BACKGROUND
Pure cloud designs centralize compute, simplifying fleet management but incurring WAN latency
and egress fees. Edge designs deploy inference, filtering, or aggregation near data sources
(gateways, routers, single board computers), reducing payload sizes and improving
responsiveness for control loops such as machine vision based quality checks or industrial
safety shutoffs.
EVALUATION APPROACH
A practical comparison logs end to end latency, packet loss, and cost under realistic traffic.
1) Establish a baseline with cloud only processing; 2) Deploy an edge preprocessor that filters
and down samples sensor frames; 3) Measure 95th percentile latency and monthly egress for
identical workloads; 4) Stress test with transient WAN loss to evaluate resilience.
FINDINGS (ILLUSTRATIVE)
In a pilot with 100 cameras streaming 1080p at 5 FPS, edge JPEG down sampling (to 480p) and
motion triggered upload reduced egress by ~72% while keeping inference accuracy within 1 3% of
cloud only results. Median action latency dropped from ~420 ms (cloud) to ~120 ms
(edge assisted). Cost savings concentrated in reduced egress and smaller cloud GPU footprints.
DESIGN CONSIDERATIONS
Use containers for repeatable deployment, an over the air update channel, and device
attestation. Prefer stateless edge services with idempotent cloud ingestion. Implement
back pressure and local buffering for intermittent connectivity.
SECURITY AND SAFETY
Harden devices (secure boot, encrypted storage), rotate credentials, and restrict services by
default. Log locally and mirror to cloud when online. For safety critical control, ensure
fail safe states and watchdogs.
1
LIMITATIONS AND FUTURE WORK
Edge fleets add operational complexity hardware variance, environmental wear, and distributed
debugging. Future work includes standardized observability for edge nodes and scheduling that
jointly optimizes power, thermals, and model accuracy.
CONCLUSION
Edge assisted architectures are compelling for latency sensitive IoT. Small pilots with careful
observability can quantify benefits before scaling.
REFERENCES
1) Shi et al., "Edge Computing: Vision and Challenges"; 2) Satyanarayanan, "The Emergence of