[22/07/2022] Experts at the Institute for Computer Systems Architecture, in the School of Informatics, in collaboration with their colleagues world-wide, have developed vHive, an open ecosystem for benchmarking and experimentation across the next-generation cloud infrastructure. The initiative aims to join the efforts of the broad industrial and academic community (experts and researchers from more than twenty leading universities and five global software and hardware vendors) to enable joint innovation, equip academic and competing industrial researchers with cutting-edge technologies and set a concrete benchmarking standard in the cloud industry. Image The future of cloud computing is serverless Any modern business in a modern digital economy increasingly relies on online services deployed in cloud datacenters to operate. The new-generation cloud, called serverless, boosts developer productivity, allowing the developers to focus on their service’s business logic rather than infrastructure details. However, this cloud architecture leads to a fundamental disconnect between the developers and cloud vendors, where the former are agnostic to the internals of the cloud infrastructure whereas the latter lack the knowledge of the online services. Although allowing faster time to market for the end products, the disconnect between the vendors and the developers results in poor service user experience and low energy efficiency of cloud datacenters. Cloud vendors (such as Amazon Web Services and Microsoft Azure) operate large-scale cloud datacenters, allowing external IT professionals to rent cloud compute resources to run online services on top of the vendors’ commercial cloud offerings. Cloud vendors keep their infrastructure proprietary, to protect their intellectual property from their competitors, preventing external IT professionals from optimising their workloads for the specifics of the underlying infrastructure. At the same time, the cloud vendors lack understanding of the internal characteristics of the online services, developed by the businesses. In serverless, developers can focus on writing solely their application code while providers take responsibility for scaling each function’s resources according to traffic changes. This labour division opens up great opportunities for systems researchers who seek to innovate in serverless computing. Unfortunately, leading serverless providers, like Amazon Web Services (AWS) Lambda and Microsoft Azure Functions, continue to rely on proprietary infrastructure, leaving academics to black-box research with production offerings or building their in-house prototypes of individual serverless infrastructure components. vHive ecosystem enables joint academia and industry innovation To enable computer systems research in serverless clouds, the experts at the University of Edinburgh’s Institute for Computer Systems Architecture (ICSA) have devised vHive, an open ecosystem for serverless benchmarking and experimentation, with the goal of enabling researchers to study and innovate across the entire serverless software stack. Building on top of state-of-the-art open-source software projects, vHive integrates industry-leading technologies efficiently in an end-to-end architecture representative of cutting-edge commercial cloud offerings, offering similar high performance and high information security guarantees. The vHive initiative is led by Dr. Dmitrii Ustiugov and Prof. Boris Grot and the EASE lab at ICSA. vHive drives innovation in many domains of computer systems, including benchmarking methodologies, operating systems, and communication networks. For benchmarking, the researchers have devised STeLLAR, a cloud-agnostic benchmarking framework, which allows to analyse and compare the performance characteristics of proprietary infrastructure of different commercial clouds. The ecosystem features a framework for prototyping complete end-to-end systems, which are representative of commercial serverless clouds, and includes vSwarm, a diverse suite of serverless benchmarks. The framework and benchmarks together allow to tailor the cloud infrastructure to running existing online services as well as exploring avenues for expanding the application domain of serverless computing to emerging services that are currently infeasible to run in serverless clouds. Equipped with the vHive ecosystem, ICSA researchers have identified a number of problems fundamental to the modern architecture of serverless clouds. The researchers demonstrated that current clouds exhibit poor performance in the presence of internet traffic changes and are ill-suited for modern datacentre hardware, compared to conventional, previous-generation cloud services, due the fundamental differences in workload characteristics. In particular, frequent changes in the internet traffic result in significant response time increase for the services running in serverless clouds, which work poorly for serving sporadic queries coming from the end service users. To improve user experience, the researchers suggested a technique for launching new instances of serverless functions on demand and presented an approach for server hardware specialisation that improves hardware efficiency in the presence of many short-running functions contending for hardware resources. To date, the vHive ecosystem has already received contributions from several universities (Stanford, EPFL, ETH, TU Munich) and international companies (Huawei research centres in Munich and Edinburgh as well as Instrumental in the USA) and is in use by more than 25 organisations world-wide. vHive has been recognised with a Distinguished Artifact Award at ASPLOS 2021, a top computer systems conference. Thanks to the connection with the industry, vHive has demonstrated knowledge transfer from academia to industry leaders, with the REAP technique support added in Amazon Web Services’ Firecracker hypervisor v1.1. There is large room for improving energy efficiency and performance of today’s clouds. At ICSA, we have ongoing projects in increasing visibility into proprietary cloud infrastructure and the underlying technologies as well as exploring new approaches to design and orchestrate serverless infrastructure to help the research community to radically improve the efficiency of the current cloud systems. Dr Dmitrii Ustiugov, lead researcher This story has been sourced by Adarsh Doddappagouda Patil from the Informatics Science Communications Group and developed by the researchers at ICSA. Related links Link to Dmitrii Ustiugov's personal page Link to Boris Grot's personal page Link to Firecracker Hypervision v1.1 on GitHub This article was published on 2024-03-18
