北京时间 3 月 21 日星期二 9:00 AM - 9:50 AM
Accelerating Massive Route Planning for Food Delivery with GPU
Online food-delivery platforms are expanding their choices, allowing customers to order from a wide variety of restaurants. Meituan owns one of the world's largest multi-person, multi-points, real-time intelligent delivery platforms. Route planning is its most critical part regarding distribution efficiency, cost-saving, and user experience. As an industrial-level technology, route planning has to be fast enough. But it's a combinatorial optimization problem: this is a hard non-deterministic polynomial time (NP) problem, which is different from normal deep learning problems. We'll show you how to implement a GPU computing tens of thousands of route-planning tasks based on a two-stage, fast heuristic algorithm proposed by Meituan. We continuously optimized this application with NVIDIA's support, and achieved high enough performance to complete 6,400 parallel route-planning tasks in hundreds of millisecond with single GPU.