×
Community Blog Function Compute Helps LinkedCare Tap into AI-assisted Dental Care

Function Compute Helps LinkedCare Tap into AI-assisted Dental Care

The article introduces how LinkedCare utilized Function Compute to enhance the efficiency of its orthodontic treatment algorithms.

By Jiang Wei, Gong Yang, Zhou Tao, and Wang Bin

Background

Founded in 2015 and headquartered in Shanghai, LinkedCare has established more than 20 branches across China and owns 100 software copyrights, 91 registered trademarks and 35 patents. LinkedCare is committed to providing integrated operation and management solutions for consumer healthcare enterprises, such as dental clinics and medical aesthetic institutions.

As an industry leader in dental healthcare, LinkedCare strives for excellence in both technology and service. It provides orthodontic treatment algorithms for dental clinics, allowing dentists to take photos of visitors' oral and dental conditions and use algorithms to offer appropriate diagnoses and treatment suggestions. This improves the efficiency of dentists' work.

Platform Characteristics and Business Pain Points

Dental clinics usually operate from 8:30 to 18:00, and expect more visitors on holidays. The business exhibits obvious peaks and valleys in activity over the course of a month.

LinkedCare purchased a batch of GPU servers in the cloud earlier to deploy algorithms and provide external services, but encountered several issues in use as follows:

Low resource utilization and waste: Servers are billed on a monthly basis, and LinkedCare owns multiple servers out of concern for occasional peaks. However, there is no business traffic during off-hours, resulting in low utilization during off-peak periods. Overall resource utilization remained around 5%.

Slow service response during peak periods: The ability to handle unexpected traffic spikes is inadequate. When traffic spikes unexpectedly, the service load increases and the request response time gets longer, which directly affects the end-user experience. During peak hours, there were instances where a single request had to queue for 10 minutes.

Inadequate monitoring and troubleshooting capabilities: Orthodontic treatment algorithms are continuously iterated, and occasional stuttering and program errors occur during iterations. However, due to inadequate monitoring and alerting capabilities, these incidents were not detected promptly and were only recognized upon feedback from end stores. This reduces the efficiency of proactive algorithm optimization.

In addition, frequent O&M and continuous platform-building activities brought a heavy workload to O&M staff and increased system instability. Given the rapid development of GPU technologies in recent years, LinkedCare staff also have to dedicate more effort to this area. Faced with all these issues, LinkedCare staff are looking for a better and more efficient cloud-based solution.

Solution

In order to find a better solution to reduce costs and optimize service experience and O&M efficiency, the technical team of LinkedCare compared several products of Alibaba Cloud and finally chose Function Compute.

Alibaba Cloud Function Compute is a fully managed event-driven compute service. It allows users to write and upload code without the need to manage infrastructure resources such as servers. Function Compute intelligently allocates computing resources, runs code in an elastic and reliable manner, and provides features such as log query, performance monitoring, and alerting.

Besides traditional CPU computing power, Function Compute also supports GPU computing power. It uses serverless computing to provide on-demand GPU computing resources for users. Compared with resident GPU computing resources, serverless GPUs improve resource utilization and elasticity while reducing costs. By providing a more convenient and efficient GPU computing service, Function Compute can effectively handle accelerated workloads such as AI model inference and training, accelerated audio and video creation, and graphics and image acceleration.

The following figure shows the architecture of the GPU resources in Function Compute.

1

Function Compute adopts virtualization technology for GPU-accelerated instances to achieve strong isolation of computing power, GPU memory, and faults and full compatibility with native applications. To ensure fast delivery of computing power, Function Compute provides two tiers of resource pools. The GPU resource pool is owned by Function Compute, and users can purchase resources as needed without paying for idle resources.

The following figure shows the resource request model in GPU-accelerated instances of Function Compute.

2

After a GPU function is deployed, users can configure an auto scaling policy to allocate provisioned GPU-accelerated instances. The instances provide the infrastructure that is required for real-time inference scenarios. Function Compute performs Horizontal Pod Autoscaling (HPA) for provisioned GPU-accelerated instances based on customer-defined metrics. Requests are preferentially sent to the provisioned GPU-accelerated instances for inference. Function Compute eliminates cold starts and allows services to run at low latency. Furthermore, Function Compute integrates observability, logging, monitoring, alerting, and other systems to simplify troubleshooting and routine O&M.

Finally, after a series of validations, the technical team of LinkedCare decided to use the following architecture in Function Compute.

3

The architecture is quite simple and the business process is as follows:

  1. A customer makes an orthodontic treatment algorithm into a standard image and uploads the image to Alibaba Cloud Container Registry (ACR).
  2. When the frontend sends a request to call the orthodontic treatment algorithm and the instance needs to be initialized, Function Compute pulls the image in ACR and the underlying GPU resources to initialize the instance and deploy the algorithm application.
  3. The orthodontic calculation request is sent to the newly created GPU application for calculation and the result is returned.

Achievements and Benefits

By deploying the GPU compute load in Function Compute, the technical team of LinkedCare has solved the aforementioned issues:

Cost optimization: Function Compute provides the pay-as-you-go billing method to charge fees based on the actual request processing time. This reduces the cost of holding GPU resources by approximately 90% compared to purchasing GPU resources on a monthly basis.

Service experience during peak periods: By making resource provisions in advance of business peaks and scaling resources on demand during unexpected traffic spikes, backend resources are delivered in a timely manner. Since the deployment of Function Compute, end stores no longer experience long queues, which greatly improves the user experience.

Efficient O&M: The built-in monitoring, logging, and alerting systems of Function Compute enable monitoring of key business indicators in real time. By configuring the monitoring and alerting settings, users can receive notifications immediately after exceptions occur. With the help of the comprehensive logging system and the professional technical support team of Function Compute, errors can be located and solved quickly.

Moreover, the deployment of Function Compute ensures excellent scalability for the entire business system, eliminating concerns about core GPU resource allocation for future business growth and laying a solid foundation for sustainable business development.

Summary and Outlook

By migrating core applications to Function Compute, the technical team of LinkedCare has successfully addressed the challenges of business growth, significantly optimized the cost structure, and accelerated the development and O&M processes, achieving unprecedented agility and efficiency.

Looking ahead, the technical team of LinkedCare will continue to deepen collaboration with Function Compute. As the company's business continues to expand, more scenarios are expected to benefit from the auto scaling, cost-effectiveness, and high efficiency provided by Function Compute. The technical team of LinkedCare plans to prioritize Function Compute in new business deployments to further speed up time to market, improve user experience, and continuously reduce operational costs.

The Function Compute team also looks forward to working with LinkedCare to explore more efficient and intelligent medical service solutions to facilitate the digital transformation of the healthcare industry. We believe that by working closely with Function Compute, LinkedCare can better serve patients and medical practitioners, helping the industry move toward greater intelligence and efficiency.

0 1 0
Share on

Alibaba Cloud Serverless

100 posts | 7 followers

You may also like

Comments

Alibaba Cloud Serverless

100 posts | 7 followers

Related Products