Stability Assurance Solutions for an Unattended Parking Cloud SaaS Platform

Background

With the widespread adoption of online payments, the parking industry has seen new internet‑based services such as online reservations, coupons, self‑service payment, and frictionless fee collection. Traditional locally hosted, non‑networked servers can no longer meet these demands. Since 2017, the "TongTong Parking" team has been developing a parking‑cloud SaaS solution, moving from trial in 2018 to full rollout in 2020, and now serves thousands of online parking lots processing tens of millions of vehicle records daily.

System Architecture

Challenges in Real‑World Scenarios

3.1 Project‑Side Network Latency and Jitter

Parking lots often share internet and office networks to reduce costs, leading to short‑term latency and jitter between parking devices and the cloud, causing service interruptions.

3.2 High Real‑Time Passage Requirements

During peak hours, the system must complete user verification, billing logic, and device control within 1–2 seconds; any delay results in unacceptable traffic congestion.

3.3 High Data Integrity Requirements

Because parking involves large cash and online payment transactions, all abnormal passage data must be 100% uploaded to the cloud for financial reconciliation and audit.

Solution

4.1 Direct Hardware‑to‑Cloud Connection

To reduce unreliable intermediaries, devices connect directly to the cloud using a lightweight MQTT protocol, which offers low overhead, bidirectional communication, and QoS 1 reliability.

Publish/subscribe decouples applications.

Only 2‑byte header reduces traffic.

Supports bidirectional communication.

Provides QoS 1 delivery guarantee.

Data flow diagram:

Data Separation

Business data (small, latency‑sensitive) and file data (large, latency‑tolerant) are sent asynchronously to optimize performance.

Multiple Retransmissions

Each MQTT message carries a unique serial number and a response feedback mechanism; if the sender does not receive an acknowledgment, the message is retransmitted, improving reliability on weak networks.

Step‑wise Upload

When a device experiences prolonged disconnection, it caches unuploaded passage records locally and uploads them via a separate "step upload" interface once connectivity is restored, ensuring data completeness.

Process Supervision

A lightweight Go‑based supervisor (go supervisord) runs on devices, offering cross‑platform binaries, rich process configuration, comprehensive logging, and a simple web GUI to keep services alive and up‑to‑date.

No external dependencies; supports Linux, Windows, macOS, Arm32/Arm64.

Extensive process configuration options.

Detailed start/monitor logs for post‑mortem analysis.

Lightweight web GUI.

4.2 Adoption of Go Language Stack

Go was chosen for its cross‑platform compatibility, built‑in garbage collection, small static binaries, rich standard library, and lightweight goroutine model, simplifying concurrent development on embedded devices.

4.3 Dual‑Network Backup

Parking sites use a primary wired network and a secondary 4G wireless router; traffic normally flows over wired, automatically switching to 4G on failure and reverting when the wired link recovers.

4.4 Dual Backup for Billing Clients

During long‑term wired outages, the system can switch from a desktop billing client to a mobile billing app, then back when connectivity is restored.

4.5 Pre‑Release Model Driven Edge Solution

To mitigate network instability, business logic is computed in the cloud, and qualifying release data is synchronized to edge devices for local, sub‑200 ms decision making, effectively a "cloud compute, edge store" collaboration.

Conclusion and Outlook

The development of the parking cloud revealed that no single technology can guarantee stability; each potential failure point must be reinforced. Future work will enhance IoT platform capabilities such as remote configuration, OTA upgrades, anomaly reporting, and device‑side log collection to further support unattended parking scenarios.