Shulou(Shulou.com)06/02 Report--

The 2019 China SDN/NFV Conference, sponsored by the SDN/NFV/AI Standards and Industry Promotion Committee, was held at Crowne Plaza Hotel, New Yunnan, Beijing, April 17-18. Huang Shuquan, Technical Director of Kyushu Cloud, was invited to attend the meeting and shared the keynote speech "StarlingX-based Edge Computing Machine Learning Optimization" at the Multi-access Edge Computing Sub-Forum, and demonstrated the real-time face recognition scene based on StarlingX architecture to the attendees.

Huang Shuquan is the technical director of Kyushu Cloud and has been working on open source software for more than 10 years. Since 2011, he has been actively involved in the development contribution of OpenStack, serving as a technical contributor, speaker and chairman. In 2018, Huang Shuquan became the first Chinese technical member of the StarlingX Technical steering Committee. Kyushu Cloud is also one of the founding members of the StarlingX project.

SDN applications that combine machine learning and edge computing bring new experiences and opportunities to a variety of industries, such as real-time prediction in monitoring, self-driving cars in cars, and so on. However, building and training ML models requires a lot of resources, which is not suitable for edge. And reasoning requires much less resources, which is usually done in real time when new data is available.

In this speech, Huang Shuquan shares how to optimize the recognition task by putting it only on the edge. Use StarlingX to build a central cloud for model training and an edge cloud for identification. By leveraging the distributed cloud nature of StarlingX, models built and trained in the central cloud can be deployed to the edge cloud, and reasoning can be run locally on the edge cloud using connected devices such as cameras and accelerators. Huang Shuquan showed a real-time face recognition demonstration based on StarlingX architecture on the edge. The following is an overview of the speech:

What is edge computing?

With the emergence of new applications such as AR, VR, self-driving, Internet of things and video intelligent analysis, as well as the upcoming 5G era, the rise of edge computing has been promoted. From my own understanding, edge computing is actually an extension of cloud computing.

Edge computing is a distributed open platform that integrates the core competencies of network, computing, storage and application on the edge side of the network close to the object or data source, and provides edge intelligent services nearby. to meet the key needs of industry digital in agile connection, real-time business, data optimization, application intelligence, security and privacy protection.

It can be used as a bridge between the physical and digital worlds, enabling smart assets, intelligent gateways, intelligent systems, and intelligent services. Edge computing can provide low latency and high response services that cloud computing cannot provide, such as edge analysis, edge security firewall and so on.

What factors drive the development of edge computing

I think there are four major factors driving the development of edge computing, including latency, bandwidth, security, and linking. Many problems of cloud computing are solved through OpenStack, while edge computing helps us to generate computing near the user or near the data end, which also requires computing, network, storage and other resources.

Cloud computing also manages computing, storage and network infrastructure. Edge computing and cloud computing have a lot in common in solving problems, but they are also very different.

The data center may have a higher requirement in dustproof, temperature and so on, but the environment of edge computing is relatively bad, and edge computing will take into account some unattended and high availability.

Introduction to StarlingX

StarlingX is an open source edge computing and Internet of things cloud platform optimized for high-performance and low-latency applications. The project is supported by the OpenStack Foundation, a new top-level OpenStack Foundation pilot project. It provides a scalable and reliable edge infrastructure that has been tested and can be used as a complete stack. Its applications include industrial Internet of things, telecommunications, video transmission and other ultra-low latency areas. StarlingX ensures compatibility between various open source components and provides unique project components for fault management and service management to ensure high availability of user applications. StarlingX is the foundation for edge computing in scalable solutions and is ready for deployment.

StarlingX inherits some of the latest practices of OpenStack in the field of cloud computing and optimizes it with some specific problems we have found in edge computing. Therefore, from the technical line, it is not difficult to find that edge computing is not completely separated from cloud computing, it conforms to the development trend of cloud computing technology, and even produces higher requirements.

I am honored to join the TSC Technical Committee at the beginning of StarlingX. The main members of StarlingX TSC are from Intel, Fenghe, Huawei, Ericsson and Interdynamx.

What problem did StarlingX solve?

I just mentioned that StarlingX is based on OpenStack. What problem does it solve? I think there are four points:

The architecture of edge computing infrastructure is different from traditional cloud computing: traditional cloud computing is in the computer room, and the scale of traditional cloud computing is relatively large, while the environment of edge computing is relatively bad, for example, it is deployed at the telecommunications office. It often requires only one server to run a complete architecture to manage the network. It requires a lighter deployment, or a single deployment.

Edge computing has higher requirements for stability and robustness than traditional cloud computing: with the deployment of edge computing nodes, there may be thousands of edge nodes, and if it is deployed in our COS in the future, its number will increase sharply, and edge computing software needs higher functions such as unattended and automatic recovery.

The management of large-scale edge nodes is different from traditional cloud computing: it is more complex and difficult.

The interaction and cooperation between the edge cloud and the central cloud: it is necessary to do a good job in the coordination between the edge node and the edge node, and between the edge node and the center node.

The goal of StarlingX

Apply proven cloud technology to edge computing, and then develop a management framework for edge computing to simplify the deployment of edge cloud. Finally, it is applied to transportation, energy, manufacturing, retail, video, smart cities, self-driving, health care and other fields. It is a vision of StarlingX that we orchestrate all the resources between the central cloud and the edge cloud through edge computing as a whole.

StarlingX edge virtual platform

StarlingX provides an edge cloud software platform that is flexible to deploy, easy to scale, and highly available. Currently, StarlingX mainly includes the following new features:

Configuration management: StarlingX code provides node configuration and inventory management services, with automatic discovery and configuration of new nodes, which are critical to deployment and manage a large number of remote or inaccessible sites. Horizon's graphical user interface and a command-line interface manage the inventory of CPU,GPU, memory, large memory pages, and encrypted / compressed pages.

Fault management: users can set, clear or query alarms and logs customized for important events, either on infrastructure nodes or on virtual resources such as virtual machines and networks. Users can access the active alarm list (Active Alarm List) and active alarm counting panel (Active Alarm Counts Banner) on the graphical user interface of Horizon.

Host management: StarlingX software provides lifecycle management functions to manage hosts through a REST API interface. This vendor-independent tool detects host failures and initiates automatic repair by providing monitoring and alerts for cluster connections, critical process failures, resource utilization thresholds, and hardware failures. The tool is also connected to the motherboard management controller for auxiliary reset (out-of-band reset), power on / off, and hardware sensor monitoring, and shares host status with other StarlingX components. "

Service management: the life cycle management of providing services is realized by providing high availability based on N + M or N redundancy models across multiple nodes. This service supports the use of multiple messaging paths to avoid split-brain communication failures and the use of active or passive monitoring to clearly define the impact of service failures through a fully data-driven architecture.

Software management: this service allows users to deploy updates to correct content and new features using a consistent mechanism for all infrastructure stacks from the kernel to OpenStack services. This module completes rolling upgrades, including parallelization and support for host reboots, allowing workloads to be removed from the node by using live migration.

Three deployment models supported by StarlingX

These are the three deployment models supported by StarlingX, with scalability from small to large.

Single Server: runs all services and supports stand-alone deployment

Dual Server: redundant design

Multiple Server: fully flexible, autonomous allocation, supporting multi-node deployment

StarlingX next Generation Container Architecture

After this year's Open Infrastructure Denver Summit, a new version of StarlingX will be released, which will deploy OpenStack on K8S based on containerized deployment. Will also integrate K8S and so on:

The development direction of StarlingX: OpenStack containerization, deployment on Kubernetes clusters, and OpenStack-Helm-based management of cluster lifecycle.

Integrate Kubernetes:Docker Runtime, Calico CNI plugin and CEPH as persistent storage backend, HELM as package management, and local Docker image repository.

Other containerized edge application deployment is supported.

Marginal machine learning Demo

Next, we show a real-time face recognition demo based on StarlingX architecture at the edge of the scene.

One bare metal starts two virtual machines to deploy a highly available central cloud. Another bare metal deploys a set of Simplex edge clouds. Start the openvino virtual machine on the edge cloud, download the openvino image from the central cloud, and create the virtual machine using the openvino image on the edge cloud. In the aspect of edge virtual machine upgrade, upload the upgraded openvino image to the central cloud, and use the upgraded openvino image to rebuild the virtual machine in the edge cloud.

First of all, the logic flow is the edge box recognition and docking camera, then the edge cloud runs the basic face recognition, pulls other AI computing function models from the central cloud and loads them to run. Finally, the edge application processes the image data collected by the camera, and carries on the AI recognition in the model.

At the booth of this conference, Kyushu Cloud staff showed the real-time face recognition demo based on StarlingX architecture to the guests, which became a highlight of the whole conference and attracted many professionals to stop.

At present, Kyushu Cloud has actively participated in the work of Akraino Edge Stack community, OpenStack Foundation Edge Computing working Group, Edge Computing Industry Alliance and SDN/NFV Industry Alliance, and entered into the field of edge computing in an all-round way. Kyushu Cloud hopes to work with members of the alliance to make the edge computing ecology gradually prosperous, become the key force to support the landing of edge computing, and deepen the digital transformation of the industry.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.