Data-Aware Task Allocation for Achieving Low Latency in Collaborative Edge Computing

Sahni, Yuvraj; Cao, Jiannong; Yang, Lei

doi:10.1109/jiot.2018.2886757

Cited by 104 publications

(61 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Other works focus on methods how the tasks within a network of collaborative edge devices should be distributed [6,22]. However, these works handle general tasks and in contrast to our work focus on the network parameters.…”

Section: Related Workmentioning

confidence: 96%

DeeperThings: Fully Distributed CNN Inference on Resource-Constrained Edge Devices

Stahl

Hoffman

Mueller-Gritschneder

et al. 2021

Int J Parallel Prog

View full text Add to dashboard Cite

Performing inference of Convolutional Neural Networks (CNNs) on Internet of Things (IoT) edge devices ensures both privacy of input data and possible run time reductions when compared to a cloud solution. As most edge devices are memory- and compute-constrained, they cannot store and execute complex CNNs. Partitioning and distributing layer information across multiple edge devices to reduce the amount of computation and data on each device presents a solution to this problem. In this article, we propose DeeperThings, an approach that supports a full distribution of CNN inference tasks by partitioning fully-connected as well as both feature- and weight-intensive convolutional layers. Additionally, we jointly optimize memory, computation and communication demands. This is achieved using techniques to combine both feature and weight partitioning with a communication-aware layer fusion method, enabling holistic optimization across layers. For a given number of edge devices, the schemes are applied jointly using Integer Linear Programming (ILP) formulations to minimize data exchanged between devices, to optimize run times and to find the entire model’s minimal memory footprint. Experimental results from a real-world hardware setup running four different CNN models confirm that the scheme is able to evenly balance the memory footprint between devices. For six devices on 100 Mbit/s connections the integration of layer fusion additionally leads to a reduction of communication demands by up to 28.8%. This results in run time speed-up of the inference task by up to 1.52x compared to layer partitioning without fusing.

show abstract

Section: Related Workmentioning

confidence: 96%

DeeperThings: Fully Distributed CNN Inference on Resource-Constrained Edge Devices

Stahl

Hoffman

Mueller-Gritschneder

et al. 2021

Int J Parallel Prog

View full text Add to dashboard Cite

show abstract

“…In particular, it develops feedback functions for detecting the status of edge servers since the resource contention problem arises in hotspot edge servers. In MSGA [17], the data location and the congestion in networks are considered when transferring tasks from IoT devices to edge servers. If networks are congested, searching a detour mechanism is triggered based on the past history of flow states.…”

Section: B Task Scheduling In Edge Computingmentioning

confidence: 99%

“…The core of the collaborative task scheduling scheme is two heuristic algorithms, Completion time-based Task Assignment (CTA) and Hierarchical Weight Allocation (HWA), used in the task distributor and the task executor, respectively. The CTA algorithm, whenever a new task is generated, selects the target device (i.e., a device in which the task will be executed) and sends it to the device as soon as possible, unlike prior studies ( [17], [18], [20], [24], [41]) in which tasks are aggregated during an epoch and distributed to devices simultaneously at the end of the epoch. This real time task distribution is particularly effective to time-critical tasks.…”

Section: Collaborative Task Schedulingmentioning

confidence: 99%

Collaborative Task Scheduling for IoT-Assisted Edge Computing

et al. 2020

View full text Add to dashboard Cite

The Internet of Things (IoT) is evolving rapidly and requires IoT devices to have more resources to meet the growing needs in diverse application domains. Despite increasing demands, modern IoT devices do not fully utilize somewhat over-provisioned computing resources. In this work, we introduce a new concept of IoT-assisted Edge Computing which makes use of consolidated idle resources in IoT devices for edge services through offloading edge tasks to nearby IoT devices. For the IoT-assisted edge computing be beneficent, two important conditions should be satisfied: 1) offloaded edge tasks to IoT devices do not hurt normal execution of local IoT tasks, and 2) computing resources in IoT devices should be effectively exploited to increase the throughput of edge services. To that end, we propose a collaborative task scheduling for IoT-assisted edge computing, in which an edge node determines where to offload edge tasks among participating IoT devices based on the offloaded execution time and energy consumption, and each IoT device determines when to execute the offloaded tasks considering local tasks execution. Experimental results show that the proposed scheme not only achieves near-optimal task throughput but also outperforms other scheduling algorithms in terms of deadline satisfaction ratio of time critical tasks, while guaranteeing deadlines of local tasks in IoT devices. INDEX TERMS Edge service, IoT-assisted edge computing, IoT service, task scheduling.

show abstract

“…In a heterogeneous distributed shared‐memory multiprocessor system (HDSMS), the processing time and energy consumption of the same task on different processors are different, and access time and energy consumption of transmitting the same data between processors and memories are also different 4,5 . References 4,5 studied TSDA on a HDSMS, and they have been received extensive attention 8‐14 . In this work, we also investigate the TSDA problem based on HDSMS.…”

Section: Introductionmentioning

confidence: 99%

Task scheduling on heterogeneous multiprocessor systems through coherent data allocation

Deng

Shen

Cao

et al. 2021

Concurrency and Computation

View full text Add to dashboard Cite

Energy consumption has become one of the main bottlenecks that limit the performance improvement of heterogeneous multiprocessor systems. In a heterogeneous distributed shared‐memory multiprocessor system (HDSMS), each processor can access all the memories, and each data can be stored in different memories. This article aims at addressing the problem of task scheduling and data allocation (TSDA) on HDSMS. To minimize the total energy consumption under a time constraint for TSDA, we propose two algorithms: the extended tree assignment for task scheduling incorporating data allocation (ETATS‐DA) and critical path task scheduling and data allocation (CPTSDA). The ETATS‐DA algorithm first utilizes the extended tree assignment to search the near optimal solution for task assignment, and then allocates data to memory based on the result of assignment. The CPTSDA algorithm considers TSDA jointly on a critical path simultaneously. Our proposed algorithms perform coherent data allocation under the consideration of best task scheduling by running two different heuristic strategies, respectively, and taking the best result as the final result. We conduct a large number of simulation experiments to test the performance of our algorithms, and the results validate the higher performance of our methods compared with the state‐of‐the‐art algorithms.

show abstract

Data-Aware Task Allocation for Achieving Low Latency in Collaborative Edge Computing

Cited by 104 publications

References 32 publications

DeeperThings: Fully Distributed CNN Inference on Resource-Constrained Edge Devices

DeeperThings: Fully Distributed CNN Inference on Resource-Constrained Edge Devices

Collaborative Task Scheduling for IoT-Assisted Edge Computing

Task scheduling on heterogeneous multiprocessor systems through coherent data allocation

Contact Info

Product

Resources

About