Source-linked AI summary
Challenges and Opportunities in Edge Computing
Blesson Varghese, Nan Wang, Sakil Barbhuiya, Peter Kilpatrick, Dimitrios S. Nikolopoulos
TL;DR
Cloud-centric processing places increasing demands on communication and computational infrastructure, while edge devices have limited resources and distant-cloud analytics can impose latency and energy costs. The paper defines edge computing as moving selected workloads toward network edge nodes and surveys the challenges and opportunities involved. It concludes that edge computing remains in its infancy but has significant potential for more efficient distributed computing.
Problem
Centralized cloud processing creates latency, energy, traffic, and resource pressures, while a framework for real-time computation across heterogeneous edge nodes is not yet available.
Method
The paper defines edge computing and surveys five research challenges and five opportunities for realizing computation on edge nodes.
Results
The paper concludes that edge computing is still in its infancy but has significant potential to enable more efficient distributed computing.
Takeaways & Limitations
Realizing edge computing requires research and coordination across edge-node resources, lightweight processing, standards, benchmarking, security, and industry–academic collaboration.
Takeaways & Limitations
Front-end devices remain constrained by hardware and middleware limitations, restricting complex local analytics and potentially draining their batteries.
Abstract
from arXiv · showhide
Many cloud-based applications employ a data centre as a central server to process data that is generated by edge devices, such as smartphones, tablets and wearables. This model places ever increasing demands on communication and computational infrastructure with inevitable adverse effect on Quality-of-Service and Experience. The concept of Edge Computing is predicated on moving some of this computational load towards the edge of the network to harness computational capabilities that are currently untapped in edge nodes, such as base stations, routers and switches. This position paper considers the challenges and opportunities that arise out of this new direction in the computing landscape.
I. INTRODUCTION
Edge computing moves selected computation from distant centralized cloud data centres toward edge nodes near data sources to address latency and infrastructure demands. The paper frames this shift through its motivation, challenges, and research opportunities.
- Motivation: Cloud centralization increases communication between edge devices and geographically distant data centres, limiting applications requiring real-time responses.Edge devices include smartphones, tablets, wearables, and gadgets.
- Motivation: Edge computing explores computation on routers, switches, and base stations through which network traffic is directed.These devices are termed edge nodes and can complement cloud computation.
- Motivation: The introduction identifies five needs motivating computation on edge nodes, including distributed placement closer to data sources.The paper presents these motivations alongside challenges and opportunities in edge computing.
- Motivation: Latency-sensitive applications can require 25ms to 50ms responses, while a Canberra–Berkeley cloud round trip is approximately 175ms.The comparison illustrates why distant-cloud analytics poses a latency challenge.
- Motivation: Edge nodes one hop from users can complement cloud computations and reduce network latency for applications such as video streaming and on-demand gaming.The proposed locations include routers and base stations.
2) Surmounting Resource Limitations of Front-end Devices
Front-end devices have limited hardware and middleware resources, restricting complex analytics and potentially making local processing costly in battery consumption.
- 2) Surmounting Resource Limitations of Front-end Devices: Front-end devices capture text, audio, video, touch, and motion data but cannot generally perform complex analytics because of hardware and middleware limitations.They often send data to the cloud for processing and receive meaningful information in return.
- 2) Surmounting Resource Limitations of Front-end Devices: Local complex analytics may be possible at the expense of draining the device battery.This constraint motivates sending computational workloads to cloud or nearby infrastructure.
- 2) Surmounting Resource Limitations of Front-end Devices: Not all front-end data needs to be used to construct analytical workloads on the cloud.The passage points toward filtering or analyzing data before cloud processing, though the text ends before completing that proposal.
3) Sustainable Energy Consumption
Growing cloud workloads raise concerns about data-centre energy consumption. The paper identifies edge processing as a possible way to perform some analytics nearer the data source.
- 3) Sustainable Energy Consumption: Data centres in the next decade are likely to consume three times as much energy as today.The projected increase motivates energy-efficient strategies.
- 3) Sustainable Energy Consumption: Performing some analytical tasks on base stations or routers could alleviate increasing data-centre energy demands in a small proportion.The passage qualifies this as a limited potential alleviation rather than a complete solution.
- 3) Sustainable Energy Consumption: Edge processing may avoid overloading data centres with trivial tasks that could be performed near the data source without significant energy implications.This is presented as a sensible power-management strategy.
4) Dealing with Data Explosion and Network Traffic
Rapid growth in edge devices and generated data increases pressure on data centres and central network paths. Hierarchical computation and nearby edge nodes are proposed to distribute processing and traffic.
- 4) Dealing with Data Explosion and Network Traffic: One-third of the world’s population was expected to have a smartphone by 2018, while 43 trillion gigabytes of data were anticipated for 2020.The projections illustrate the scale of device and data growth discussed in the paper.
- 4) Dealing with Data Explosion and Network Traffic: Analytics on individual edge devices is restricted by resource limitations and cannot realistically support collective analytics from multiple devices.This limits the ability of devices alone to address growing data volumes.
- 4) Dealing with Data Explosion and Network Traffic: Nodes a hop away in the network could complement device or data-centre computation while distributing traffic and coping with data growth.The passage identifies edge nodes as an intermediate computational location.
- 4) Dealing with Data Explosion and Network Traffic: A hierarchical pipeline may filter data on the device, execute analytics on edge nodes, and reserve more complex tasks for the cloud.Alternative arrangements include cloud offloading to edge nodes and edge nodes using volunteer devices.
- 4) Dealing with Data Explosion and Network Traffic: Edge-computing frameworks remain unavailable, and real-time edge processing, workload placement, connection policies, and node heterogeneity remain open research areas.Existing cloud frameworks support data-intensive applications but do not resolve real-time processing at the network edge.
Challenge 1 - General Purpose Computing on Edge Nodes
Edge computing can use nodes between edge devices and the cloud, but many existing nodes are specialized and may not support general-purpose analytical workloads. Addressing this requires discovery, rapid benchmarking, upgrades, virtualization, or substantial hardware investment.
- Challenge 1 - General Purpose Computing on Edge Nodes: Edge computing can use access points, base stations, gateways, routers, switches, and other nodes between devices and the cloud.
- Challenge 1 - General Purpose Computing on Edge Nodes: Base stations may be unsuitable for analytics because their specialized DSPs are not designed for general-purpose computing.It is also unclear whether they can perform additional computations alongside existing workloads.
- Challenge 1 - General Purpose Computing on Edge Nodes: General-purpose capability can be added through router upgrades, virtualization, or replacing DSPs with CPUs.Replacing specialized DSPs requires a huge investment.
- Challenge 1 - General Purpose Computing on Edge Nodes: Edge-node discovery and benchmarking must handle many heterogeneous devices, generations, and modern workloads such as large-scale machine learning.Benchmarking must rapidly reveal resource availability and capability.
Challenge 3 - Partitioning and Offloading Tasks
Edge offloading requires automated task partitioning that can place computation across cloud and edge locations without requiring users to specify every node capability or location.
- Challenge 3 - Partitioning and Offloading Tasks: Distributed systems partition workflows for execution across multiple geographic locations, usually through explicit languages or management tools.
- Challenge 3 - Partitioning and Offloading Tasks: Edge offloading must automate task partitioning without requiring explicit definitions of edge-node capabilities or locations.
- Challenge 3 - Partitioning and Offloading Tasks: Users may need computation pipelines arranged sequentially across the data centre and edge nodes or executed simultaneously across multiple edge nodes.
Challenge 4 - Uncompromising Quality-of-Service (QoS) and Experience (QoE)
Edge computing must preserve both node-level service quality and user experience while accommodating additional workloads. This requires capacity-aware scheduling, monitoring, and safeguards for shared infrastructure.
- Challenge 4 - Uncompromising Quality-of-Service (QoS) and Experience (QoE): QoS captures quality delivered by edge nodes, while QoE captures quality delivered to users.
- Challenge 4 - Uncompromising Quality-of-Service (QoS) and Experience (QoE): Edge nodes should not be overloaded with computationally intensive workloads, while maintaining throughput and reliability for intended workloads.
- Challenge 4 - Uncompromising Quality-of-Service (QoS) and Experience (QoE): Peak usage knowledge is needed to partition and schedule tasks flexibly, supported by management across infrastructure, platform, and application levels.The framework raises monitoring, scheduling, and rescheduling issues.
- Challenge 4 - Uncompromising Quality-of-Service (QoS) and Experience (QoE): Public edge-node deployment must address ownership risks, preserve the device’s intended purpose, and secure multi-tenancy.
IV. OPPORTUNITIES
The paper identifies research opportunities around making edge computing publicly accessible through clarified responsibilities, standards, reliable benchmarking, and marketplace mechanisms.
- IV. OPPORTUNITIES: The paper identifies five opportunities for academic research despite the challenges of realizing edge computing.
- IV. OPPORTUNITIES: Edge computing’s public accessibility depends on articulating the responsibilities, relationships, and risks of all parties.
- IV. OPPORTUNITIES: Existing cloud standards must be reconsidered to address edge-node stakeholders and the social, legal, and ethical aspects of edge use.This requires commitment and investment from public and private organizations and academic institutions.
- IV. OPPORTUNITIES: Reliable benchmarking against well-known metrics is necessary for standards implementation, while SLAs and pricing models are needed for an edge marketplace.
Opportunity 2 - Frameworks and Languages
Edge computing requires new frameworks and toolkits because existing cloud workflow models may not express user-driven edge analytics, making programming more complex than cloud-accessible models.
- Opportunity 2 - Frameworks and Languages: Edge nodes supporting general-purpose computing will require frameworks and toolkits tailored to edge analytics.
- Opportunity 2 - Frameworks and Languages: Existing workflows, developed mainly for scientific domains such as bioinformatics and astronomy, may not suit user-driven edge analytics.
- Opportunity 2 - Frameworks and Languages: Programming models for edge analytics must account for use cases that differ from established distributed scientific workflows.
- Opportunity 2 - Frameworks and Languages: Developing edge-oriented programming support is more complex than making the cloud accessible through existing models.
Opportunity 3 - Lightweight Libraries and Algorithms
Edge nodes impose hardware constraints that exclude heavyweight software, creating demand for lightweight libraries, algorithms, operating systems, and collaborative research validated against real infrastructure.
- Opportunity 3 - Lightweight Libraries and Algorithms: Edge nodes cannot support heavyweight software because their hardware resources are constrained.A cited example contrasts a 4-core ARM-based base station with Apache Spark’s stated requirement for at least 8 CPU cores and 8 GB memory.
- Opportunity 3 - Lightweight Libraries and Algorithms: Edge analytics therefore need lightweight algorithms capable of reasonable machine-learning or data-processing tasks.
- Opportunity 3 - Lightweight Libraries and Algorithms: Apache Quarks supports basic filtering and windowed aggregates on small-footprint devices but is insufficient for advanced tasks such as context-aware recommendations.
- Opportunity 3 - Lightweight Libraries and Algorithms: Micro operating systems or microkernels could reduce resource use while supporting quicker deployment, shorter boot times, and resource isolation on heterogeneous edge nodes.
- Opportunity 3 - Lightweight Libraries and Algorithms: Edge research benefits from industry and government access because many academic institutions cannot access infrastructure needed to validate and refine their work.
- Opportunity 3 - Lightweight Libraries and Algorithms: An open consortium of operators, developers, cloud providers, and academics could drive edge-computing research for mutual benefit.
V. CONCLUSION
Edge computing remains in its infancy but could enable more efficient distributed computing; the paper presents five challenges and five opportunities for realizing it.
- V. CONCLUSION: Edge computing is still in its infancy and has potential to enable more efficient distributed computing.
- V. CONCLUSION: The paper highlights five research challenges and five opportunities involved in realizing edge computing.