AVAR

Customer background

AVAR is a metaverse and web3 virtual label, with virtual human IPs, virtual fashion brands, and digital art generators under its umbrella. Its core technologies include a 3D procedural generation framework, a character fusion algorithm, and AR applications. The application scenarios cover virtual idols, virtual fashion, and digital collectibles NFT. AVAR uses procedural algorithms to generate digital assets with the largest scale, highest efficiency, and greatest influence; it aims to become a virtual label for metaverse digital asset production and creation; with algorithms driving creativity, virtual technology innovating real – world demands, and building a 3D content ecosystem in the metaverse. The core technologies are the 3D procedural generation framework, the virtual character fusion algorithm, and AR applications. The application scenarios include virtual idols, virtual fashion, virtual clothing, and digital collectibles. Currently, AVAR has established partnerships with Xiaohongshu, Alibaba Digital Collection, various variety shows, and celebrity studios.

Aiuni is an innovative platform under AVAR that focuses on the field of AI – generated content (AIGC). It aims to provide users with a fast – conversion service from 2D images to high – quality 3D models through its self – developed Unique3D technology. With the mission of “making 3D creation simpler”, Aiuni promotes the popularization and high – efficiency of 3D content creation through its efficient and intelligent AI tools. Whether it is game development, film and television production, or virtual reality (VR) and augmented reality (AR) applications, Aiuni can provide users with powerful technical support and convenient solutions.

Business requirements and challenges

With the global expansion and promotion of the AVAR business, the access frequency of the Aiuni.ai website has increased significantly. However, based on the existing Volcano Engine Cloud Platform, users generally report obvious website access delays and slow image loading speeds. These issues have led to an increase in user dissatisfaction with the Aiuni.ai platform. In view of this, AVAR urgently needs to find a cloud service provider with better network performance.

In the context of the rapid development of current artificial intelligence technology, the demand for computing resources is increasing day by day. In the existing business architecture, AVAR makes extensive use of T4 and A100 graphics cards. However, during peak business periods, some inference tasks and training tasks on the Volcano Engine platform may queue up and wait due to insufficient graphics card resources. This not only affects the task execution efficiency but also reduces customer satisfaction.

In addition, in terms of scheduling at the business front end, due to the adoption of a self – built Kubernetes (K8S) cluster, the resource utilization rate can reach as high as 180% during peak business periods, but drops to 30% during off – peak periods, resulting in a large amount of idle resources and failing to achieve the maximum utilization of resources.

In terms of data security and compliance, the challenges faced by AVAR are particularly complex and urgent. With the global expansion of the business, the compliance requirements for data storage, processing, and transmission vary by region. The compliance requirements for data storage and processing in different regions are becoming increasingly stringent. The existing platform has deficiencies in meeting multi – regional data compliance, which may bring potential legal risks.

solution

In the project, Baidao, a Google partner, assisted AVAR’s IT team in conducting a detailed analysis of the cost investment plan and optimization solutions for the cloud business on Google Cloud. At the same time, it provided very good answers and business solutions to various problems and requirements. During the transformation process, some Google Cloud products were used to help AVAR design an infrastructure based on GCP, improve the old architecture on Volcano Engine, enabling customers to shift towards automated infrastructure management in service management, focus on user experience, and maintain the stable operation of the system.

Description of the solution architecture

• High – performance computing platform

With the Compute Engine of Google Cloud, customers can carry out a large number of inference tasks and model training tasks without worrying about performance bottlenecks caused by computing power. In addition, in some business scenarios, customers can also appropriately use Spot GPU models, which not only meet the requirements of inference tasks but also significantly reduce usage costs.

At the same time, with the continuous growth of the number of users, the Aiuni platform can flexibly scale up or down resources through Google Kubernetes Engine (GKE), ensuring that the service can provide stable support during peak business periods. When demand decreases, the configuration can be automatically scaled down to save costs.

• Data storage and management

By adopting Google Cloud’s Cloud Storage service, multi-region deployment can be achieved, which not only ensures compliance with the laws and regulations of various regions but also significantly improves users’ access speed. In addition, Cloud Storage also provides a data encryption function. Customers can upload custom keys to perform content-level encryption on sensitive files, thus ensuring the security of files stored in the cloud. At the data access level, the access link is signed and verified through the signed-cookie mechanism, further ensuring the high security of data files and access control. 

• Content Distribution Service

By adopting Google Cloud’s Cloud CDN service, it is possible to provide efficient and low-latency content distribution services to global users. Cloud CDN caches static content to edge nodes distributed globally, enabling users to obtain the required content from the nearest node, thus significantly improving the content loading speed and optimizing the user experience. In addition, relying on Google Cloud’s global network infrastructure, only one global external application load balancer needs to be configured to achieve efficient content distribution for global users, ensuring that users can quickly access the required resources regardless of their location.

• Image processing service

In the actual business scenarios, it is often necessary to crop images to specific sizes. With the Cloud Run service of Google Cloud, the relevant services can be built using containerized deployment. In this mode, customers do not need to spend energy on managing the underlying infrastructure, but only need to focus on code writing and deployment. In addition, Cloud Run adopts a pay-as-you-go model, where costs are incurred only when the code is actually running, and no costs are incurred when it is not running. This feature can effectively reduce business costs and improve resource utilization efficiency.

• Service Monitoring and Management

Log Collection (Logging): Collect logs of applications and services for troubleshooting and performance analysis.

Performance Monitoring (Cloud Monitoring): Monitor the performance indicators of the monitoring platform, such as CPU usage, memory usage, request latency, etc., so as to detect problems in a timely manner and optimize them.

Cost Management (Billing API):  Track and manage the usage costs of cloud resources to control budgets and optimize resource allocation.

Use the product

• Google Kubernetes Engine
• Compute Engine
• Cloud Load Balancing
• Cloud CDN
• Cloud SQL for MySQL
• Cloud Run
• Filestore
• Artifacts Registry

Customer earnings

Enhanced Performance and Scalability: Compared to the previous infrastructure based on ByteDance Cloud, using Google Kubernetes Engine (GKE) for microservices orchestration has reduced application latency during peak hours by 30%. GKE’s excellent auto – scaling capabilities enable the platform to handle unexpected surges in demand seamlessly, ensuring a consistently responsive user experience.

Reduce infrastructure costs: After using the Spot machines of GKE, customers can maximize the utilization of resources, ensuring the stable operation of their businesses while reducing infrastructure costs by 40%.

Operation and Maintenance Management Tools: By adopting Cloud Logging and Cloud Monitoring, the entire process of log collection and analysis is simplified. Alerts for issues can be promptly sent to the customer’s communication tools, which reduces the Mean Time to Repair (MTTR) of critical incidents by 30%.