Case Study: Volcano Engine achieves 3× PB-scale data processing efficiency with LanceDB

Volcano Engine boosts PB-scale autonomous driving data processing 3x with LanceDB

Volcano Engine LAS, a service from Bytedance, faced significant challenges managing the massive, unstructured multimodal data (text, images, point clouds) generated by autonomous driving systems for a leading Chinese automotive enterprise. Their legacy storage system struggled with data explosion, high storage costs, inefficient compute scaling, and an inability to add new data columns without completely rewriting the entire dataset, which led to GPU underutilization.

LanceDB provided a solution built on the Lance format, which enabled zero-cost data evolution for dynamic labeling, transparent compression that reduced point cloud storage by 70%, and point query optimization for training. This resulted in a 3x improvement in EB-scale data processing efficiency, a 40% acceleration in model training delivery, and an increase in GPU utilization from 60% to over 90% for the customer.

Volcano Engine