Application features and differences between UNINet and OneNET

The Meaning of UNINet

UNINet stands for Unicom Net, which is the GPRS network access point of China Unicom. Through UNINet, users can obtain full Internet access rights and enjoy NAT (Network Address Translation) services. This network allows China Unicom's users to connect to the Internet via GPRS (General Packet Radio Service) technology.

 

Main Features:

• International Export Bandwidth: Up to 3,807 Mbps.

• Service Types: Including dial-up and dedicated line access, VPN/VPDN services, VISP services, WWW applications, email services, etc.

• Network Architecture: It adopts the networking mode of IP + ATM (Asynchronous Transfer Mode), with high bandwidth and large capacity characteristics.

Applications of UNINet

UNINet not only plays an important role in traditional mobile Internet access but also has a variety of innovative applications in modern computing and network fields.

1. Unified Architecture Search

In the field of deep learning, UNINet is used as a hybrid network architecture that combines Convolutional Neural Network (CNN), Transformer, and Multilayer Perceptron (MLP). This architecture models different operators in a unified form and uses the same set of searchable configuration parameters (such as operator type, expansion, channels, etc.) to describe each different operator. In addition, the proposed Context-Aware Downsampling Module (DSM) helps to coordinate the combination among different operators. In this way, UNINet shows powerful performance and efficiency improvements in visual benchmark tests.

Feature Analysis:

• Shallow Convolution, Deep Transformer: The convolutional layers are mainly used for extracting local features in the shallow layers, while the Transformer layers are placed in the deep layers to capture global dependency relationships.

• Balanced Computation Distribution: Similar amounts of FLOPs (Floating Point Operations) are distributed between the convolution and the Transformer to ensure the effective utilization of computing resources.

• Downsampling Strategy: The Local Downsampling Module (L-DSM) and the Local-Global Downsampling Module (LG-DSM) are used to adapt to the representations of the convolution and the Transformer respectively.

2. Network Representation Learning (NRL)

UNINet is also applied to the representation learning of large-scale networks. By introducing an efficient edge sampler and a random walk model abstraction, this framework can effectively perform representation learning on large-scale networks. Specifically, the edge sampler based on the Metropolis-Hastings sampling technique can quickly converge to any discrete probability distribution, thus supporting diverse random walk models. This enables UNINet to exhibit excellent efficiency when dealing with large-scale networks.

Experimental Verification:

• Performance on Large-Scale Datasets: Experiments on multiple real-world datasets show that UNINet can efficiently handle large-scale networks with billions of edges, significantly improving the learning efficiency and effect.

• Open Source Implementation: Open source code is provided to facilitate further exploration and application by researchers and developers.

3. Container Network Acceleration

In the cloud environment, UNINet is designed to accelerate the data plane of container networks. By offloading container-level network tunnels through SmartNIC (Smart Network Interface Card), UNINet improves network throughput, reduces tail latency, and significantly reduces CPU usage. This solution is especially suitable for VM-based IaaS environments and solves the performance degradation problem caused by double network tunnels.

Performance Improvement:

• Throughput: An average improvement of 7.08 times.

• Tail Latency: A reduction of 41.6%.

• CPU Usage: At the receiving end, it is reduced by up to 5.6 times, and at the sending end, it is reduced by 4.02 times.

In summary, UNINet not only provides Internet access services for China Unicom's GPRS users but also plays an important role in multiple cutting-edge technology fields such as deep learning, network representation learning, and cloud network acceleration. These applications demonstrate the strong adaptability and high efficiency of UNINet in different scenarios.

Features and Functions of OneNET

OneNET is China Mobile's Internet of Things open platform. It is an open platform and ecological environment built by China Mobile IoT Co., Ltd. based on the characteristics of the Internet of Things technology and industry. OneNET supports various network environments and protocol types, is suitable for the access of various sensors and smart hardware and the provision of big data services, and provides rich APIs and application templates to support the development of various industry applications and smart hardware. The features of OneNET include end-to-end security, a simple and easy development experience, selective data sharing, and a fully interconnected infrastructure.

Features of OneNET

As a PaaS Internet of Things open platform built by China Mobile, OneNET has the following features:

• Device Access and Connection: It helps developers easily achieve device access and connection, quickly complete product development and deployment, and provides a complete Internet of Things solution for smart hardware and smart home products.

• Support for Multiple Protocols: It supports multiple standard protocols and API development to meet the high-concurrency and rapid access requirements of a large number of devices, including protocols such as LWM2M(CoAP), MQTT, Modbus, HTTP, and TCP.

• Data Management and Analysis: It provides functions such as data storage, message distribution, capability output, event alarm, and data analysis. Based on distributed cloud storage, message object structure, and rich data call interfaces, it realizes high-concurrency read and write operations on the database and effectively ensures data security.

• Wide Range of Application Fields: The Internet of Things private network has been applied to various commercial fields such as environmental monitoring, remote meter reading, smart agriculture, smart home appliances, smart hardware, energy conservation and emission reduction, vehicle networking, industrial control, and logistics tracking.

• Usability and Extensibility: Simple APIs and documents make rapid integration possible, allowing even beginners to quickly get started. As the number of Internet of Things devices grows, the architecture of OneNET allows seamless expansion to meet the needs of large-scale deployment.

• Security: A comprehensive data encryption and permission control system ensures data security.

Functions of OneNET

The functions of OneNET mainly include the following aspects:

• Device Access: It supports the access of devices with multiple protocols and provides two product types, namely public protocols (such as EDP, MODBUS, MQTT, HTTP, etc.) and private protocols, to facilitate the access of different types of devices.

• Device Management: It provides operations such as device registration, binding, and unbinding, supports the life cycle management, online status monitoring, online debugging, and data management of devices, and also supports functions such as device synchronization commands and device mirroring.

• Data Storage: It provides data storage services, and developers can store the data collected by devices in the cloud.

• Data Analysis: It supports the analysis and processing of device data, generating reports, charts, etc., and provides unified data management and analysis capabilities based on Hadoop, etc.

• Data Visualization: It provides a graphical interface to facilitate developers to view device data.

• Message Distribution: It quickly informs business platforms, user mobile phones, and APP clients of various collected data through message forwarding, short message and multimedia message pushing, and APP information pushing, establishing an effective two-way communication channel.

• Capability Output: It gathers the core capabilities of China Mobile such as short message and multimedia message, location service, video service, and public cloud, provides standard API interfaces, and shortens the development cycle of terminals and applications.

• Event Alarm: It builds an event trigger engine, and users can quickly implement application logic orchestration based on the engine.

• Application Development Environment: It provides a SaaS application development environment to support developers in the development and management of Internet of Things applications.

Differences between UNINet and OneNET

1. Definition and Purpose

• UNINet: UNINet is the access point of China Unicom's GPRS network. Through UNINet, users can obtain full Internet access rights and it also provides NAT services. It is an important part of China Unicom's GPRS network and is mainly used to provide high-speed Internet access services.

• OneNET: OneNET is the Internet of Things platform launched by China Unicom, mainly providing connection management and services for Internet of Things devices. It supports multiple Internet of Things communication protocols, allowing users to remotely monitor and manage connected devices and realize the collection, processing, and analysis of device data.

2. Service Scope and Features

• UNINet: UNINet mainly provides high-speed Internet access services for individual users and enterprise users and is suitable for scenarios where quick access to the Internet is required. It supports the access of multiple devices, including smartphones, tablets, etc., and can provide a stable network connection speed.

• OneNET: OneNET focuses on the field of the Internet of Things and provides services such as device management, data storage, and big data analysis. It is suitable for enterprises and developers who need to remotely control and monitor a large number of Internet of Things devices, helping users achieve intelligent management and data value mining.

3. Technical Architecture and Service Model

• UNINet: The technical architecture of UNINet is based on China Unicom's GPRS network, and users' network connections are realized through specific access points. Its service model is usually charged according to the amount of data flow, and users pay fees according to the actual amount of data used.

• OneNET: The technical architecture of OneNET is more complex, involving multiple links such as the connection management of Internet of Things devices, data transmission, cloud storage, and big data processing. Its service model is diverse and can provide customized solutions according to users' needs, including free trials, paid subscriptions, and other forms.

In summary, although both UNINet and OneNET are products of China Unicom, there are obvious differences in the objects they serve, application scenarios, and technical architectures. UNINet mainly provides high-speed Internet access services, while OneNET focuses on the field of the Internet of Things and provides comprehensive device management and data services.

Service Areas of the OneNET Platform

The OneNET platform is an open cloud platform launched by China Mobile IoT Co., Ltd., aiming to provide convenient connection, cloud storage, message distribution, and big data analysis services for Internet of Things applications and industry solutions. The OneNET platform mainly provides services in the following fields:

• Smart Home: It provides device life cycle management tools to help quickly realize the cloud management of large-scale devices and the generation of customized intelligent applications.

• Smart Vehicle: It supports functions such as vehicle status monitoring and remote control, promoting the development of vehicle networking.

• Smart Wearable: It supports the collection and analysis of data from devices such as health monitoring and sports tracking.

• Smart Energy: It provides services such as energy consumption monitoring and energy efficiency analysis to help intelligent energy management.

• Industrial Manufacturing: It supports production process monitoring and equipment maintenance management, improving the automation and informatization level of the manufacturing industry.

In addition, the OneNET platform also provides rich APIs and data distribution capabilities, supports the access of devices with multiple industry standard protocols, such as NB-IoT (LWM2M), MQTT, EDP, Modbus, HTTP, etc., to meet the needs of different application scenarios. The platform also has features such as access enhancement, edge computing, value-added capabilities, AI, data analysis, one-stop development, industry capabilities, and ecological openness, and can adapt to various complex Internet of Things scenarios.

The services of the OneNET platform are not limited to the above fields, and it is constantly expanding new application scenarios to meet the digital transformation needs of more industries.

Research Progress of UniNet

UniNet is also a research achievement. It is the first work to unify convolution, Transformer, and MLP for searching. The performance of UniNet exceeds that of EfficientNet and SwinTransformer, and it achieves a top-1 accuracy of 84.9% on ImageNet, which is better than EfficientNet-B7 and BoTNet-T7.

Performance of UniNet in Image Recognition Tasks

UniNet is a new network architecture that combines Convolutional Neural Network (CNN) and Transformer, aiming to explore a unified visual recognition architecture. According to the latest research, UniNet has demonstrated good performance in image recognition tasks.

The design concept of UniNet is to improve the performance of the model by adding inductive bias. This inductive bias can help the model learn certain patterns or rules more easily. The overall architecture of UniNet still follows the common pyramid structure, which contains four hierarchical stages, and each stage includes a downsampling layer and multiple unified blocks. These unified blocks can handle features of different scales, thus improving the performance of the model. UniNet also introduces High-Dimensional Convolution (HdC), which is a lightweight 3×3 depthwise separable convolution that can perform local fusion on high-dimensional features, encode high-dimensional implicit features, and improve the efficiency of the model.

The experimental results of UniNet show that it can steadily improve the performance of all spatial token mixers and narrow the performance gap among them. In addition, UniNet equipped with naive local window attention can even surpass the previous state-of-the-art technology. All models and codes will be publicly released, which means that other researchers can verify the performance of UniNet and conduct further research based on it.

In summary, the performance of UniNet in image recognition tasks is positive. Through its innovative network architecture and algorithm design, it has demonstrated its potential in the field of visual recognition. With more research and practice, UniNet is expected to play a greater role in future image recognition tasks.

Summary

UNINet and OneNET are two different concepts. The former is the network access point and multi-service unified network platform of China Unicom, while the latter is the Internet of Things open platform of China Mobile. UniNet, on the other hand, is a new network search method that combines the advantages of multiple network architectures.