The HEIDIS project aims at designing and experimenting a new three-layer scheduling framework for 5G and beyond 5G open and programmable network environments. Besides the vertical coordination between the hierarchical three scheduling layers that has to be embedded in the algorithms, we will also work on defining the horizontal interfaces among the schedulers of multiple resources. The involved resources include the radio one, the IP link resource – via SDN systems and programmable Smart-NIC (Network Interface Cards) – and the computing systems (CPU, RAM, and storage) resources behind NFV/Cloud architecture. We will show how the three-layer view can be transposed to other resources than the radio, and how horizontal interfaces among per-resource schedulers can support the consideration of multi-resource allocation constraints needed to improve resource efficiency and multi-resource fairness. We aim at experimenting the proposed scheduling algorithms by leveraging on the Open RAN (O-RAN) platform expected to revolutionize access networks, augmented in terms of functionalities and algorithms with open Software Defined Networking (SDN) and Cloud/Network Functions Virtualization orchestration, control-plane and data-plane elements.
The algorithmic and architectural elements of the HEIDIS framework will be evaluated against two use-cases: a Smart Grid use-case, where tight coupling is needed between edge servers calculating grid reconfigurations for grid protection strategies (to anticipate demand overload) hence requiring ultra-reliable and low latency services and related resource scheduling, and an Infrastructure Sharing use-case calling for high programmability of the scheduling logic to allow the integration of differentiated service policies on a per-operator basis.
HEIDIS addresses this research field with a world-class group of experts, from both industry and academia with very complementary skills. Academic partners include experts in Cloud-RAN architecture (UPSaclay) and in network softwarization and automation (CNAM, Paris). Industrial partners include a major network operator, Orange, with its R&D department (Orange Labs) working in network softwarization and ONAP/ORAN/OPNFV systems, a power distribution operator, EDF with its R&D department working on smart grids, and SMILE, European leader in embedded systems software.
The implementation of the project is to be fulfilled by carrying out four main work packages (WPs) in parallel during the funding period:
WP0. Project management (Resp. UPSaclay): The goal of this WP is to track the progress of milestones and deliverables against the project schedule with the agreed budget/time frame.
WP1. Project Architecture and Scheduling Requirements:The goal of this WP is to set the architectural basis and the requirements for the HEIDIS scheduling framework encompassing radio, transport and computing resources, for forthcoming 5G and beyond-5G network environments undergoing function disaggregation and softwarization. The introduction of function decomposition into microservices, the increasing adoption of novel serverless computing and Function as a Service (FaaS) paradigms in the scaling of computing resources, at both radio access and core network functions is indeed exposing new opportunities for dynamic resource allocation within radio resource architecture layers and across multiple resource domains; in this context, the radio access takes a key leading functional role in the multi-resource provisioning chain. We plan to highlight what will be the current state of interfaces stemming from the Open Radio Access Network (O-RAN) and Software-Defined RAN (SD-RAN) initiatives, introducing novel scheduling points working at different time scales but to be coordinated vertically; we will then specify novel interfaces and functional elements required for handling horizontal multi-resource scheduling and orchestration in coordination with SDN and NFV/Cloud systems, highlighting the novel challenges opportunities in related link and computing system scheduling as well. New behaviors related to the derived hierarchical and disaggregated scheduling systems and decision problems therein will be modeled using offline and online optimization, including game-theoretic and stochastic modeling. Two reference use-cases of the project are defined and characterized in WP, one about smart-grid and power systems networks, and one about multi-operator infrastructure sharing; the use-cases will also be put in relationship with the functional architecture and related scheduling models of the derived decision-making problems.
WP3. Smart-NIC systems design and programming:The goal of WP3 is to continue the development of the HEIDIS scheduling environment to the SDN environment for link resource management. Indeed, in the multi-resource HEIDIS context, the link resource is managed by an SDN architecture, where the SDN controller interacts with its radio counterpart (RIC) and its NFV/cloud counterpart (VNF Manager), to perform near-real time control-plane operations in the configuration of data-plane switching elements as specified in D1.3 and developed in D2.3 for the SD-RAN interfaces.. Besides SDN switches, the design of Smart-NICs for highly programmable SDNs is a hot research topic thanks to their stronger programmability potential using stateful switching and advanced switching rule definition with languages such as P4. In HEIDIS, we plan to design a Smart-NIC system able to integrate the multi-resource hierarchical and disaggregated scheduling with real-time stateful switching behaviors. We plan to design the SDN real-time scheduler using P4-NetFPGA systems and contribute to the corresponding open-source project.
WP4. Open-Source Management and Experimentation: The goal of WP4 is the validation of the implementations of components and the proposed scheduling algorithms in an experimental platform assessing the project Use Case. The basic identified software platform components are O-RAN Software Community, ONOS, CORD/OMEC, and (if needed) ONAP for, respectively, the radio, link, NFV/MEC and orchestration layers, and other open-source initiatives that may be launched in the time frame from the submission to the actual project kick-off. Thanks to the project, this basic software ecosystem is expected to get enriched with the novel interfaces among these functional bricks defined in WP1 and designed in WP2, and the possibility to use Smart-NIC for network acceleration as studied in WP3. WP4 will also cover the open-source contribution effort to the above-mentioned existing open-source projects, as well as the launch and maintenance of novel open-source projects related to project activities. WP4 will take advantage of the test environment built around the Open Testing and Integration Center (OTIC) of the O-RAN community made available by Orange on its premises starting from fall 2021, which will be part of a European lab network. The Open Testing and Integration Centre (OTIC) provides a collaborative, open, and impartial working environment. Multiple OTICs may exist in different regions around the world; each of them may focus on different O-RAN functions, interfaces, blueprints, as well as different cloud-deployment scenarios as those depicted in Fig. 2 for VNF testing etc. In this context, Orange intends to open an OTIC on its premises at Orange Gardens-Chatillon, connected with other labs in Europe in order to provide an end-to-end O-RAN chain for test and integration. In this sense, the HEIDIS partners will be able to use the tool chain and other open source and proprietary components that will be necessary to validate the proposed architecture and scheduling algorithms.