Edge Computing for Connected Cars - Use Case, Requirement and Architecture

### Overview Toru Kurusawa, a researcher at Toyota working on edge computing technologies for connected vehicles, presented the use cases, service requirements, and architectural challenges facing connected cars. By around 2025, an estimated 100 million connected cars worldwide are expected to generate between 1 to 10 exabytes of data per month. Transmitting and concentrating all this data directly into the cloud creates an overwhelming load on both mobile networks and centralized cloud infrastructure. To manage this massive load, edge computing is positioned as an essential distributed architecture.

### Key Topics Connected car services are categorized into three primary service types, each with distinct communication requirements: - **In-Vehicle Infotainment (IVI):** Combines entertainment and information (GPS navigation, streaming music, voice apps). Prioritizes user experience and primarily relies on broadband cellular downloads. - **Intelligent Transport Systems (ITS):** Focuses on safety and vehicle control (danger warnings, collision avoidance, cooperative adaptive cruise control). Requires ultra-low latency and highly reliable communications, suitable for Dedicated Short-Range Communications (DSRC) and cellular V2X (C-V2X). - **Vehicle IoT / Vehicle Big Data:** Encompasses high-definition map data, drive recorder video, and diverse sensor inputs. It requires high capacity. While individual latency demands may be flexible, the shear volume can congest networks and disrupt latency-sensitive services like ITS.

To illustrate the scale, Kurusawa noted that if 3 million Toyota cars upload 20 GB of data monthly, they produce 60 petabytes per month. At a cloud transaction processing rate of 10 GB/s, processing this batch alone would take over two months without distributed edge architectures. To coordinate industry solutions, the Automotive Edge Computing Consortium (AECC) was founded in 2018, bringing together automotive OEMs, telecommunications operators, and cloud providers.

### Technical Details & Proof-of-Concept Experiments Edge environments face challenges because edge clouds have limited compute availability and scalability compared to centralized public clouds, while vehicle demand fluctuates sharply based on geographic events (such as accidents or sudden traffic jams). The speaker detailed two specific proof-of-concept (PoC) implementations developed to address these fluctuations:

1. **Cross-Edge Scaling (Serverless Multi-Edge):** Using Kubernetes and Knative, the team demonstrated a serverless edge infrastructure deployed across neighboring edge sites. Connected cars first access containerized application pods at the closest edge site. As request frequency increases, pods scale out locally. Once local compute limits are reached, application pods scale out onto adjacent edge sites to absorb traffic and prevent local service outages.

2. **Dynamic Upload Throttling and Traffic Control:** To prevent ingress network congestion at the edge, a dynamic load-control mechanism was tested. Using the Open Networking Foundation's (ONF) open-source COMAC project, the team emulated an LTE mobile network with two Multi-access Edge Computing (MEC) applications positioned where the serving/packet gateway connects. The first application monitors the number of connected vehicles; as connection density grows, it instructs vehicles to reduce transmission frequency and video resolution before uploading to a video recognition service. In the PoC, adding one vehicle every 40 seconds showed that incoming network traffic remained roughly constant rather than growing linearly.

### Takeaways Edge computing closer to the cloud side provides the initial scalability needed for vehicle big data, while MEC infrastructure handles localized split traffic. Crucially, the experiments demonstrated that open-source software (such as Kubernetes, Knative, and ONF COMAC) allows enterprise end-users and automotive researchers to rapidly prototype and evaluate telecom and edge solutions independently, bypassing historically slow and costly vendor-dependent setups.