2024扩展计算服务：实现统一的网络体验白皮书

Extended compute servicesfor a unified networkingexperience UNEXT White paper series Mathieu Boussard Nokia Bell Labs is pioneering the next frontier in intelligent connectivity with the creation of theUnified Networking Experience Technology (UNEXT) platform. UNEXT is an innovative networkingsolution designed to seamlessly unify diverse network and compute capabilities into a single,intelligent ecosystem. By abstracting complexity, orchestrating heterogeneous technologies andenabling end-to-end services, UNEXT will deliver a streamlined experience for stakeholders acrossthe value chain. With its foundation in cutting-edge research and decades of expertise, Nokia Bell Labs UNEXTaims to redefine how networks operate, making them more adaptive, efficient and capable ofsupporting the demands of a hyper-digital world. Realizing services end-to-end in UNEXT will require defining the abstraction, orchestration andsupport for heterogeneous network and compute capabilities across a wide range of stakeholdersand technologies. This paper describes the foundations of the resulting secure, generalizednetwork-compute continuum, which will serve as UNEXT’s execution environment. Contents Introduction Network systems have become foundational to our modern societies. They are expected to continue togain importance, evolving in nature and functional scope while incorporating an increasing number ofboth technologies and stakeholders. This will exacerbate the complexity of both running networks andconsuming digital services. UNEXT (Unified Networking Experience Technology, [Sef23]) is our vision for aunified networking experience that systemically addresses this challenge by supporting secure and trustedinteractions across stakeholders and seamless service, network and compute orchestration over theresources they each contribute to the system. To accomplish this ambitious goal, UNEXT proposes a novel approach to creating simple, secure andscalable services based on composable autonomous software agents. UNEXT decomposes the challengeinto six main sub-problems (dealt with in as many “focus areas,” Figure 1): defining the foundational designrules and autonomous agents signature and behavior (Unified System); enabling interactions betweenmodules of various stakeholders with different goals and requirements, forming de facto a decentralizedsystem (Decentralized Environments, [atk24]); supporting the flows of data and knowledge in the system(Knowledge and Data Services, [con24]); supporting autonomy in system operation and service realization(Autonomous Services); supporting orchestration of network and compute resources across space,stakeholders and technologies (Extended Compute Services); and supporting tight interaction levelsbetween networks and applications (Network-Application Symbiosis). This whitepaper describes the context, challenges and proposed approaches related to the ExtendedCompute Services (ECS) research area. ECS aims to provide UNEXT’s execution environment, both to runUNEXT core components and UNEXT application components, defining the abstraction, orchestration, andsupport for network and compute capabilities across a secure, generalized network-compute continuum. Extended Compute Services Vision: towards a generalized network and compute continuum A relatively recent trend in distributed computing has looked at the extension of the realm of cloudorchestration to multiple domains across a “cloud continuum.” While this notion has received different andevolving definitions in the literature (Moreschini & al., 2022), proposes the following definition synthesizingprior proposals: “Cloud Continuum is an extension of the traditional Cloud towards multiple entities(e.g., Edge, Fog, IoT) that provide analysis, processing, storage, and data generation capabilities.” As this definition focuses implicitly on cloud (native) technologies and leaves out the network interconnectingthe various compute domains, others have proposed to generalize to “network-compute continuum.” Anetwork-compute continuum is an abstraction layer for heterogenous, distributed (and possibly decentralized)compute and network resources that forms a uniform execution environment for applications and networkfunctions, so they do not need to care about their kind, physical location or ownership. In Extended Compute Services, we envision a generalization of the scope of orchestrated distributed computeenvironments towards a secure, generalized network-compute continuum, which spans the three followingdimensions (Figure 2). The first dimension is the position on the cloud continuum, which depicts where service components mightbe placed in the continuum formed by the central cloud, by the edge (with its various definitions), andbeyond, by the end devices themselves, here labeled as “Extreme-Edge.” Getting different tiers along sucha cloud continuum to jointly contribute to an end-to-end service with specific requirements may requirenetwor