# envd in university ## Typical Scenario - Have 2-4 machines, typically 4 cards per machine - Shared by a group of people, usually on average less than 1 card per people - The usage will be intense before the paper deadline # Core Requirement - Manage multiple machine in a centralized platform - Submit jobs, with timeout limits - "Fair scheduling", especially before the paper deadline - Machine usage metrics - No problem with GFW - Shared memory support # What we need to provide - End2end tech stack select - selection of k8s distribution - dataset management (nfs? fluid?) # Reference: - https://zhuanlan.zhihu.com/p/422462131 - https://zhuanlan.zhihu.com/p/139158539