CAREER: SHF: Enhancing Serverless Efficiency Through Microarchitectural Checkpointing

点击翻译按钮获取中文摘要

Cloud data centers are a vital piece of technology infrastructure today. Applications and websites rely deeply on their efficient operation to serve millions to billions of users. To improve the useful utilization of cloud servers and ease application development, cloud platforms have recently developed a new design paradigm known as serverless computing. New serverless applications are made of very small pieces of code, which often run for just milliseconds, in contrast to traditional applications. This research project investigates the implications of this change on microprocessors which have been historically designed for long-running applications. The project explores how to modify the cores of the microprocessor to improve their energy efficiency when running such serverless applications. Improving the efficiency of serverless computing in cloud data centers will enable cost savings and a reduced environmental footprint across economic sectors while lowering the barrier to entry for new entrants with novel applications. In addition, this project integrates an education plan featuring a new change of major pipeline and participation in equitable and inclusive teaching practices workshops to broadly enhance the diversity of the growing cloud computing and computer science workforce.This project investigates the impacts of emerging, short-lived applications like serverless computing on the prediction structures embedded into microprocessors designed with long-lived applications in mind. Serverless applications waste significant energy and time running on today's server processors, but they have been adopted to improve overall server utilization and developer productivity. Saving the contents, or microarchitectural checkpointing, of structures like branch predictors and caches thus has the potential to drastically improve the efficiency of serverless and supercharge traditional applications by eliminating microarchitectural training time. This project investigates which structures are best to checkpoint, implements checkpointing mechanisms into commercially-adopted, open-source processors, designs the user and operating system software interfaces to the programmer, and proposes new security protections needed to make the checkpointing adoptable. This full-stack prototyping process, exploiting real-world, open-source software and hardware, will demonstrate the significant efficiency benefits realizable with microarchitectural checkpointing. This will enable more efficient data centers for serverless edge and cloud computing that are proliferating as well as traditional, long-running applications, all of which underpin the cloud services that society relies on.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.