{"id":616008,"date":"2019-10-17T15:35:49","date_gmt":"2019-10-17T22:35:49","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-project&p=616008"},"modified":"2023-06-06T15:41:13","modified_gmt":"2023-06-06T22:41:13","slug":"leap","status":"publish","type":"msr-project","link":"https:\/\/www.microsoft.com\/en-us\/research\/project\/leap\/","title":{"rendered":"LEAP: Lean, Efficient, And Predictable Cloud Platforms"},"content":{"rendered":"

No public cloud can host large latency-sensitive services, such as search engines, in a way that is economic for those services today! Project LEAP<\/strong> (short for Lean, Efficient, And Predictable) addresses the research challenges in enabling cloud platforms to host these services in a rightsized, elastic, and with predictable tail latency.\u00a0 The challenges include how to prevent interference from co-located workloads, how to prevent performance jitter due to I\/O, how to make the services elastic despite their large storage footprints, and how to leverage spare capacity to run their batch workloads without affecting any co-located workloads.\u00a0 LEAP leverages emerging hardware and sophisticated software techniques to overcome these challenges.\u00a0 As a by-product of the project, we are also introducing an increasingly popular benchmark suite of representative Azure workloads.<\/p>\n

As a concrete target, we aim to host Bing head services on Azure with similar tail latency to their current latency on bare metal. We are pursuing this target in collaboration with several product groups, including Bing, Azure, and Windows\/Hyper-V.\u00a0 The benefits are significant, including new revenue for Azure as it will host latency-sensitive services for 1st and 3rd parties; new revenue for Bing as it will monetize its internal services, such as IndexServe, by providing them as Azure services to external users; and unifying AutoPilot and Azure into a single infrastructure.<\/p>\n

Another major concrete target is to deploy aggressive resource (CPU, memory, etc) oversubscription and harvesting in Azure without producing performance impact for workloads.\u00a0 This will produce substantially lower costs for Azure.<\/p>\n

Research challenges and our current progress:<\/p>\n