In-network Computation Research Articles

A Survey on Architectures, Hardware Acceleration and Challenges for In-Network Computing

A Survey on Architectures, Hardware Acceleration and Challenges for In-Network Computing

When In-Network Computing Meets Distributed Machine Learning

When In-Network Computing Meets Distributed Machine Learning

IN3: A Framework for In-Network Computation of Neural Networks in the Programmable Data Plane

IN3: A Framework for In-Network Computation of Neural Networks in the Programmable Data Plane

SOAR: Minimizing Network Utilization Cost With Bounded In-Network Computing

SOAR: Minimizing Network Utilization Cost With Bounded In-Network Computing

Evaluating the impact of microservice-centric computations in internet of vehicles

This paper presents the design and implementation of a hierarchical and multilayered computing framework, specifically tailored to analyze the benefits of Named Data Networking (NDN) enabled microservice-centric in-network computations in autonomous vehicular networks. The proposed framework comprises three layers: an autonomous vehicular network layer, an Edge computing layer, and a centralized Cloud.Modifications were made to the vanilla NDN codebase to facilitate communication between the vehicular network layer, Edge computing and Cloud layer, utilizing the C++ Boost Asio library and IP tunneling mechanism. A microservices-based driver assistance application, consisting of various heterogeneous microservices, was emulated in .NET Core and deployed on different Edge computing terminals and the Cloud. Additionally, RESTful APIs have been developed to enable the vehicular network layer and the physical Edge layer to offload microservice-centric computation requests and retrieve the corresponding computational outcomes. The Entity framework is employed to ensure proper tracking and management of these requests. An HTML-based user interface has also been developed for a visual representation of the request pattern. Extensive testbed experiments reveal that the proposed system significantly optimizes bandwidth consumption, reduces latency, and increases the computed results delivery ratio when compared to conventional monolithic systems.

In-Network Computing Empowered Mobile Edge Offloading Architecture for Internet of Things

In-Network Computing Empowered Mobile Edge Offloading Architecture for Internet of Things

Information-Centric Function Chaining for ICN-Based In-Network Computing in the Beyond 5G/6G Era

Information-Centric Function Chaining for ICN-Based In-Network Computing in the Beyond 5G/6G Era

1.63 pJ/SOP Neuromorphic Processor With Integrated Partial Sum Routers for In-Network Computing

1.63 pJ/SOP Neuromorphic Processor With Integrated Partial Sum Routers for In-Network Computing

DINC: Toward Distributed In-Network Computing

In-network computing provides significant performance benefits, load reduction, and power savings. Still, an in-network service's functionality is strictly limited to a single hardware device. Research has focused on enabling on-device functionality, with limited consideration to distributed in-network computing. This paper explores the applicability of distributed computing to in-network computing. We present DINC, a framework enabling distributed in-network computing, generating deployment strategies, overcoming resource constraints and providing functionality guarantees across a network. It uses multi-objective optimization to provide a deployment strategy, slicing P4 programs accordingly. DINC was evaluated using seven different workloads on both data center and wide-area network topologies, demonstrating feasibility and scalability, providing efficient distribution plans within seconds.

A secure and effective data aggregation in WSN for improved security and data privacy

The increasing prevalence of internet usage and mobile devices has underscored the critical importance of safeguarding personal data. This is especially important in Wireless Sensor Networks (WSNs), where information typically requires in-network computing and collaborative processing. These computationally demanding approaches are not suitable for resource-constrained WSN nodes. Aggregating data effectively while protecting user data is a major challenge in wireless sensor networks. Many privacies of homomorphism encryption-based WSN data aggregation methods have been created and investigated recently. Since cluster leaders (aggregators) may rapidly combine cypher texts without decryption, communication overhead is reduced, making these data aggregation methods more secure than traditional ones. However, the base station only receives aggregated output, causing issues. Initial limits apply to aggregating functions. If the aggregated output is the sum of sensing data, the base station cannot acquire the maximum value. Second, attaching message digests or signatures to sensory samples does not allow the base station to validate data authenticity. In dangerous places, WSNs must be energy-efficient and private. In this research, we present a data aggregation method known as Energy-Efficient and Privacy-Preserving (E2P2). E2P2 data aggregation utilizes less energy and yields more accurate results. Private data aggregation with increased accuracy and hybrid encryption is presented in this research. The goal is to reduce data transmission and energy use collisions and offset collision-induced loss. Extensive simulations compare E2P2 to earlier approaches. Experimental results show that E2P2 outperforms other algorithms. Good exactness, complexity, and safety are demonstrated by theoretical and simulation results.

In-Network Computation for Large-Scale Federated Learning Over Wireless Edge Networks

Most conventional Federated Learning (FL) models are using a star network topology where all users aggregate their local models at a single server (e.g., a cloud server). That causes significant overhead in terms of both communications and computing at the server, delaying the training process, especially for large scale FL systems with straggling nodes. This paper proposes a novel edge network architecture that enables decentralizing the model aggregation process at the server, thereby significantly reducing the training delay for the whole FL network. Specifically, we design a highly-effective in-network computation framework (INC) consisting of a user scheduling mechanism, an in-network aggregation process (INA) which is designed for both primal- and primal-dual methods in distributed machine learning problems, and a network routing algorithm with theoretical performance bounds. The in-network aggregation process, which is implemented at edge nodes and cloud node, can adapt two typical methods to allow edge networks to effectively solve the distributed machine learning problems. Under the proposed INA, we then formulate a joint routing and resource optimization problem, aiming to minimize the aggregation latency. The problem turns out to be NP-hard, and thus we propose a polynomial time routing algorithm which can achieve near optimal performance with a theoretical bound. Simulation results showed that the proposed algorithm can achieve more than 99<inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> of the optimal solution and reduce the FL training latency, up to 5.6 times w.r.t other baselines. The proposed INC framework can not only help reduce the FL training latency but also significantly decrease cloud&#x2019;s traffic and computing overhead. By embedding the computing/aggregation tasks at the edge nodes and leveraging the multi-layer edge-network architecture, the INC framework can liberate FL from the star topology to enable large-scale FL.

DNA-Storage in Future Communication Networks

5G shifted the paradigm from mere communication to in-network computing, so that networks have started applying the compute-and-forward paradigm. With the advent of the current vision of future communication networks (e.g., 6G), in-network intelligence also has become pivotal. Data analytics and their storage for continuous accurate learning and intelligent decisionmaking have become part of the communication. This explosion of data to be stored for unknown duration of time is going to stress the future communication infrastructure, which will be unable to achieve the targeted energy usage together with low-latency and reliability. This article presents a possible solution via data storage based on synthetic DNA. We further propose approaches for the implementation of DNA data storage into future communication networks. The focus is especially on the impact that DNA storage can have for storing the historic data of a network controlled by artificial intelligence for network automation and prediction of future network states (intelligent management). In this sense, DNA data storage can offer a unique opportunity to store efficiently and reliably large amounts of data, mined during network monitoring and traffic engineering.

FoggyEdge: An Information-Centric Computation Offloading and Management Framework for Edge-Based Vehicular Fog Computing

Recent advances aiming to enable in-network service provisioning are empowering a plethora of smart infrastructure developments, including smart cities and intelligent transportation systems (ITSs). Although edge computing, in conjunction with roadside units, appears to be a promising technology for proximate service computations, the rising demands for ubiquitous computing and ultralow-latency requirements from consumer vehicles are challenging the adoption of ITSs. Vehicular fog (VF) computing, which extends the fog computing paradigm in vehicular networks by utilizing either parked or moving vehicles for computations, has the potential to further reduce computation offloading transmission costs. Therefore, with a precise objective of reducing latency and delivering proximate service computations, we integrate VF computing with roadside edge computing and propose a four-layer framework named FoggyEdge. The FoggyEdge framework is built at the top of named data networking and employs microservices to perform in-network computations and offloading. A real-world Simulation of Urban Mobility-based preliminary performance comparison validates FoggyEdge’s effectiveness. Finally, a few future research directions on incentive mechanisms, security and privacy, optimal VF location, and load balancing are summarized.

Offloading Machine Learning to Programmable Data Planes: A Systematic Survey

The demand for machine learning (ML) has increased significantly in recent decades, enabling several applications, such as speech recognition, computer vision, and recommendation engines. As applications become more sophisticated, the models trained become more complex while also increasing the amount of data used for training. Several domain-specific techniques can be helpful to scale machine learning to large amounts of data and more complex models. Among the methods employed, of particular interest is offloading machine learning functionality to the network infrastructure, which is enabled by the use of emerging programmable data plane hardware, such as SmartNICs and programmable switches. As such, offloading machine learning to programmable network hardware has attracted considerable attention from the research community in the last few years. This survey presents a study of programmable data planes applied to machine learning, also highlighting how in-network computing is helping to speed up machine learning applications. In this article, we provide various concepts and propose a taxonomy to classify existing research. Next, we systematically review the literature that offloads machine learning functionality to programmable data plane devices, classifying it based on our proposed taxonomy. Finally, we discuss open challenges in the field and suggest directions for future research.

Accelerating Distributed Cloud Storage Systems with In-Network Computing

Accelerating Distributed Cloud Storage Systems with In-Network Computing

Hybrid in-network computing and distributed learning for large-scale data processing

Hybrid in-network computing and distributed learning for large-scale data processing

An Architecture for Provisioning In-Network Computing-Enabled Slices for Holographic Applications in Next-Generation Networks

Applications such as holographic concerts are now emerging. However, their provisioning remains highly challenging. Requirements such as high bandwidth and ultra-low latency are still very challenging for the current network infrastructure. In-network computing (INC) is an emerging paradigm that enables the distribution of computing tasks across the network instead of computing on servers outside the network. It aims at tackling these two challenges. This article advocates the use of the INC paradigm to tackle holographic applications' high bandwidth and low latency challenges instead of the edge computing paradigm that has been used so far. Slicing brings flexibility to next-generation networks by enabling the deployment of applications/verticals with different requirements on the same network infrastructure. We propose an architecture that enables the provisioning of INC-enabled slices for holographic-type application deployment. The architecture is validated through a proof of concept and extensive simulations. Our experimental results show that INC significantly outperforms edge computing when it comes to these two key challenges. In addition, low jitter was maintained to preserve the hologram's stability.

Accelerating Industrial IoT Acoustic Data Separation With In-Network Computing

Acoustic data from the Industrial Internet of Things (IIoT) are widely used in anomaly detection because audio information reflects richer internal statuses of monitored working machines than the video does. Since multiple acoustic data sources interfere with each other by nature, source data estimation is a prerequisite of subsequent anomaly detection. Existing schemes often use a centralized manner to separate full data on a remote node in clouds. However, such a centralized manner may delay reactions to anomalies due to data transmission delay and the complexity of solving data separation problems. This article shows that the data separation phase can be substantially accelerated with an in-network computing approach. The key idea is to offload data processing jobs to intermediate network nodes along the forwarding path. We first propose a distributed algorithm so that the data separation jobs can be done in a progressive manner; likewise, we modify the forwarding layer in order to eliminate hop-by-hop data transmission delay that hurts the performance of using in-network computing. We further derive theoretical upper and lower bounds of the required number of intermediate nodes that achieve the maximum acceleration. We also implement our proposed solution in a full-stack network emulator. Based on an open and professional data set, evaluation results justify the feasibility and advantages of our idea with nearly 32.18% acceleration on total processing time. This work exemplifies the convergence of IIoT, edge, and clouds.

Stateful InREC: Stateful In-Network Real Number Computation With Recursive Functions

The current generation of Reconfigurable Match-Action Tables switches are highly programmable, able to support stateful operations and pipeline specifications using languages like P4. Nevertheless, these switches do not offer primitives to support real-valued operations on the data plane, thus requiring support from external servers or middle boxes to perform advanced operations. We introduce Stateful InREC, a system that extends the capabilities of programmable switches to support in-network real-valued operations using the IEEE half-precision floating point representation. Stateful InREC relies on decomposing real-valued functions into lookup tables taking into account the RMT model constraints to reach the right trade-off between accuracy and resource usage. It also supports state management for the computation of recursive function over time series. Stateful InREC prototype on Barefoot Tofino switches demonstrates the efficiency of Stateful InREC for in-network computation of different types of operations and its application for in-network logistic regression models used for classification problems. We also demonstrate the use of Stateful InREC to implement an ARIMA model on a Tofino switch for DDoS detection. Our evaluation of Stateful InREC shows that it is possible to implement complex in-network applications with high accuracy and low latency.

P4 and NetFPGA-Based Secure In-Network Computing Architecture for AI-Enabled Industrial Internet of Things

This article proposes a secure in-network computing system based on a simple reduced instruction set architecture, which can be used for processing artificial intelligence and machine learning models in network devices, in an AI-based Industrial Internet of Things (IoT) system. The architecture exploits the capabilities of upcoming generations of packet processing pipelines in programmable network switches. This instruction set enables processing of data at multiple terabits-per-second, which is beyond the processing power of current servers. Instructions for regular expressions, basic arithmetic, and logical operations are defined as a proof of concept. A packet containing both instruction and data blocks is presented as an input to the pipeline by bundling both the function and its arguments into the packet. The primary challenge in opening up network switches for executing a user-defined code is security. In this context, this article presents a secure execution model (SEM), which provides additional levels of security by deliberately disallowing memory allocation and modifications to persistent state of the network switch. Furthermore, real-life use cases are presented in this article to demonstrate the utility of the proposed instruction set architecture, as also applicable to IoT data processing. This instruction set is implemented in the programming protocol-independent packet processors language, verified on a mininet-based software switch and demonstrated on Xilinx NetFPGA SUME boards. The performance results show line rate packet processing with zero packet loss, at 10 Gb/s, and average packet latency of 3.66 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${\mu }\text{s}$ </tex-math></inline-formula> .