Haili Zhang
Cloud Platform Director of UISEE©, a leading autonomous driving company.
Cloud Native focuses: Kubernetes, DevOps, Observability, Service Mesh, Serverless.
Haili Zhang
Cloud Platform Director of UISEE©, a leading autonomous driving company.
Cloud Native focuses: Kubernetes, DevOps, Observability, Service Mesh, Serverless.
CNCF Survey 2021
- Container Adoption and Kubernetes have truly gone mainstream – usage has risen across organizations globally, particularly in large businesses.
- According to CNCF’s respondents, 96% of organizations are either using or evaluating Kubernetes – a record high since our surveys began in 2016.
- Kubernetes has demonstrated impressive growth over the past 12 months with 5.6 million developers using Kubernetes today.
Serverless Boosts 🚀
"As the popularity of FaaS products like AWS Lambda continues to grow, we have also seen a significant increase in adoption of other types of serverless technologies offered by Azure, Google Cloud, and AWS."
Serverless Platform War?
"Hosted platforms (75%) are the most popular"
Why need cloud agnostic Serverless platform?
Kubernetes brings the possibility of cloud-agnostic: multiple, distributed cloud.
But it’s difficult to be cloud-agnostic for Serverless
- Each cloud provider has its own Serverless platform
- Further more, these platforms are coupled with their own cloud backend services
Is it possible to build a cloud agnostic Serverless platform? 🤔
What is Serverless?
Let’s make core concepts clear first.
"Serverless computing = FaaS + BaaS." [1]
- Cloud functions — packaged as FaaS (Function as a Service) offerings — represent the core of serverless computing
- Cloud platforms also provide specialized serverless frameworks that cater to specific app requirements as BaaS (Backend as a Service) offerings
How to construct?
We need adapt various backend services.
Dapr decouples the distributed apps with underlying backend services.
Dapr provides you with APIs that abstract away the complexity of common challenges developers encounter regularly when building distributed applications. These API building blocks can be leveraged as the need arises - use one, several or all to develop your application faster and deliver your solution on time.
How to construct?
We need achieve "zero-scale".
KEDA is a Kubernetes-based Event Driven Autoscaler. With KEDA, you can drive the scaling of any container in Kubernetes based on the number of events needing to be processed.
- KEDA is a single-purpose and lightweight component that can be added into any Kubernetes cluster.
- KEDA works alongside standard Kubernetes components like the Horizontal Pod Autoscaler and can extend functionality without overwriting or duplication.
OpenFunction is a cloud-native open source FaaS (Function as a Service) platform aiming to let you focus on your business logic without having to maintain the underlying runtime environment and infrastructure.
You can concentrate on developing business-related source code in the form of functions.
Use Case: Online Data Processing
Using OpenFunction in Autonomous Driving
Why need a cloud agnostic Serverless platform?
Why Cloud Agnostic
- Different customers require to use different cloud providers
- Some customers’ vehicle data is sensitive and are required to be isolated from public clouds
Why Serverless
- Autonomous driving has so many application scenarios, there’re requirements to use different processing logics for different scenarios even for the same data source
- Data types and processing modules are complex, multi-language support is required
- Massive events need to be handled in real time
- Data processing logics tend to be changed frequently based on rapidly changing requirements
Dapr Matters
- Data from different sensors and processing modules· are complex
- Different cloud providers have different backend services that may result in duplicated implementations over the same data processing logic
- Pub/Sub and Bindings make it easy to set up event-driven architecture for data processing with async functions
KEDA Matters
- AI drivers need to retain human working habits to a certain extent, so there are obvious peaks and valleys in traffic based on working schedules
- For security reasons, data refresh frequently and sensors produce massive data, so workload should be scalable
- The usage of compute resources directly affects project cost