Background information
SKG is an R&D enterprise that is specialized in developing high-end wearable health devices. SKG combines fashion and technology to provide beautifully-designed massage products that promote a healthier lifestyle.
With the rapid changes in market demands, the IT system of SKG faces multiple issues such as the inaccurate inventory, isolated online and offline channels, inflexible deployment architecture, and slow response speed. To keep up with the expansion of the market, improve response efficiency, and perform large-scale business operations, SKG has formed a strategic partnership with Alibaba Cloud to build a channel-oriented mid-end project that covers all channel applications based on the connection between online and offline channels. The channel-oriented mid-end is built to support the rapid development of business by migrating business data to the cloud. The channel-oriented mid-end is applicable to the marketing management of SKG channels, such as the online channel, offline channel, and gift channel, interconnects the business data of multiple terminals, such as the distributors, shopping guides, and backend SAP, and integrates the membership data of customers and the sales data of retailers.
The marketing operation management platform, distributor portal, and shopping guide mini program that are built based on the underlying channel-oriented mid-end require the multi-channel connection capability. In addition, different channels have different custom business requirements and access methods. In this case, the channel-oriented mid-end must provide the following capabilities: flexible customization, high extensibility, and high scalability that adapt to the traffic of different channels.
Challenges
Before the channel-oriented mid-end was built, SKG rented a data center to deploy most of its applications, and ran some business on the cloud. SKG used a hybrid cloud architecture that integrated a data center, Elastic Compute Service (ECS) instances, and ApsaraDB RDS to deploy applications. The architecture requires a large number of manual O&M operations, such as releasing applications, building and importing open source resources, integrating cloud services, and managing clusters. The following table describes the challenges faced by SKG when SKG used the original system to deliver and maintain services. The table also describes the benefits of the channel-oriented mid-end that is provided by SAE.
Challenge | Benefit |
Low maturity of agile collaboration and DevOps process: Full lifecycle management is not provided during project iteration. Offline communication is the main method that is used to follow up issues and task progresses. The original system does not provide an online tracking tool. No automated tool is provided during the DevOps process. For example, you must manually release applications, the release process is inefficient and requires a long period of time to complete, and online security failures often occur.
| You can focus on designing and building applications without the need to manage and maintain clusters and servers. SAE helps you manage the lifecycle of the applications and maximize resource utilization. SAE also provides basic monitoring and application monitoring capabilities. |
Complex application release and deployment: Before you release an application, you must perform multiple operations such as evaluating resources, purchasing an application server, and installing and initializing supporting software. You must configure multiple settings, such as cluster monitoring, service governance of release and scheduling scripts, configuration management, and log backup. You can configure the settings by using independent supporting components or systems.
| You can package an image or upload a code package to deploy an application and configure canary release in the SAE console. You do not need to write scripts or log on to a server. |
Difficulties in getting started with the container technology: Troubleshooting is inefficient because developers are not familiar with the underlying container management platform and the settings are configured in a black box manner. | You do not need to purchase IaaS-layer resources such as ECS instances. You can select the CPU and memory specifications in the SAE console and configure network settings with a few clicks. |
Low scalability: During off-peak hours, the resource utilization is low. If you want to perform a scale-out operation, a release process is required and servers cannot be removed. If the channel-oriented mid-end is released, more microservices applications will be removed. The volume of traffic that is received by an application varies based on the scenario. In this case, flexible scaling policies are required.
| SAE supports fast scaling. You can perform scaling operations in the SAE console. SAE provides multiple types of auto scaling policies to handle burst traffic. The auto scaling feature allows you to adjust the CPU and memory resources in a fine-grained manner and improve resource utilization. |
High O&M costs: You must maintain applications, the infrastructure, and the corresponding supporting systems. Excessive manpower is required.
| SAE is a maintenance-free service. No O&M engineers are required. Developers can manage resources in the SAE console. |
Solution
The following figure shows the solution that is used by SKG. 
Applications are automatically deployed to SAE by using the continuous integration and continuous deployment (CI/CD) pipelines that are provided by Dayu. The channel-oriented mid-end improves the efficiency of application deployment by replacing the manual deployment method with an automatic deployment method and reducing the number of manual O&M operations. The channel-oriented mid-end also reduces the risks during a change order release, implements controllable deployment, and ensures production security. Before SKG used SAE, at least two dedicated O&M engineers were required to maintain the clusters and applications of the similar scale. After SKG used SAE, the number of O&M operations and the O&M costs are reduced. Developers can complete the configurations in the SAE console. Before SKG used SAE, the SKG team had to troubleshoot the issues that are related to cluster maintenance and the underlying checks of Kubernetes clusters and ECS instances. After SKG used SAE, the issues can be ignored.
SAE is compatible with applications that are developed based on microservices frameworks such as Spring Boot, Spring Cloud, and Dubbo. SAE also integrates with cloud services such as Application Configuration Management (ACM) and Application Real-Time Monitoring Service (ARMS). SAE blocks the underlying details to allow SKG to deploy applications with a few clicks. SAE provides flexible auto scaling policies to help SKG optimize resources.
Effect
During the implementation of the channel-oriented mid-end project, SKG uses Dayu Technology Delivery Platform to perform online delivery. The platform enables SKG to manage resources in a centralized manner. The SKG channel-oriented mid-end already deployed more than 20 applications that belong to the front-end and mid-end to SAE, such as the microservices gateway, microservices center, front-end portal, terminal mini program, and front-end node. When SKG releases an application, SKG does not need to modify the system to adapt to the configurations. SKG needs to only configure the required settings in the SAE console, and the application can stably run after the release.
SAE metrics
Only 2 to 3 hours are required to initialize, create, and deploy more than 20 SKG applications to SAE. Compared with purchasing servers, the resource costs are reduced by 30%. The resource overheads of development and test environments can be reduced by 50% because SAE provides an instance type of 0.5-core CPU. Scale-out operations can be performed within minutes, not days.
Dayu metrics
On the Dayu platform, SKG has already delivered more than 20 applications, submitted more than 1.8 million lines of code, and used pipelines to automatically release applications more than 3,000 times. The average release time is less than 100 seconds. The efficiency of application deployment is increased by 300% thanks to the CI/CD automatic deployment method. No release failures have occurred.
Elastic Compute Service (ECS)
Container Compute Service (ACS)
