How Coinbase Achieved Rapid Auto Scaling with MongoDB скачать в хорошем качестве

How Coinbase Achieved Rapid Auto Scaling with MongoDB

Learn why more than 50,700 customers have chosen to build their applications on MongoDB here: https://trymongodb.com/4d42UfS Subscribe to MongoDB YouTube→ https://trymongodb.com/4dTnbFR Crypto markets are volatile. To handle large swings in traffic Coinbase needed a way to be able to rapidly scale our fleet of MongoDB clusters up and down based on traffic. In this talk we discussed how Coinbase does capacity planning, how we performed autoscaling, and how Atlas has made vertical scaling so much faster. Speaker: Sean Hurley, Staff Engineer, Coinbase ⏱️ Timestamps ⏱️ Introduction and Background 00:00:00 Sean Hurley, a staff engineer at Coinbase, introduces himself and discusses the company's various crypto products and features. He provides specific background on the hundreds of MongoDB clusters powering different experiences at Coinbase, each with vastly different resource requirements. Challenges in Crypto Traffic and Initial Scaling Approach 00:05:09 Sean explains the volatility and unpredictability of traffic in the crypto space, highlighting the challenges they face with traffic spikes, such as those caused by Elon Musk's tweets. He describes their initial scaling approach and its limitations. Identifying Clusters for Scaling and Resource Modeling 00:10:18 The focus shifts to identifying which clusters need scaling and determining the necessary resources. Sean talks about modeling based on previous traffic spikes and correlating resource utilization with user activity to predict future needs. Traffic Prediction Model and Scaling Up in Advance 00:15:30 Sean introduces their machine learning traffic prediction model, which uses cryptocurrency price volatility to predict traffic spikes. He demonstrates how this model allowed them to scale up in advance of a significant traffic increase, ensuring a smooth user experience. Atlas Team's Approach to Improve Scaling 00:20:44 Matteo Visini Hydrick, a lead engineer on the Atlas dedicated team, discusses the partnership with Coinbase and the technical details of the improvements made to speed up vertical scaling. He outlines the two phases of improvements and their impact on scaling times. Results and Future of Cluster Scaling 00:25:52 Matteo concludes by sharing the results of their scaling improvements, which have made vertical database scaling much faster and more predictable. He also touches on the future focus for cluster scaling, including predictability and speed enhancements, and invites feedback for unique scaling scenarios.