End-to-End Orchestration of NextG Media Services over the Distributed Compute Continuum

TL;DR

This work addresses end-to-end orchestration of NextG media services over distributed compute continua by modeling information-aware processing graphs as DAGs and embedding them in cloud-network graphs. It introduces CNFlow as a framework to capture flow chaining, scaling, and replication and presents IDAGO, the first polynomial-time, multi-criteria approximation for unsplittable multicast IA-DAG-DTR, enabled by a novel DAG-to-Forest transformation. The approach yields substantial cost and latency improvements over VNE-based and information-unaware baselines, with rigorous probabilistic guarantees and extensibility to discrete resource blocks and dynamic service conditions. The results across synthetic and realistic VR scenarios demonstrate practical impact for scalable, efficient orchestration of NextG services in cloud-edge continua.

Abstract

NextG (5G and beyond) networks, through the increasing integration of cloud/edge computing technologies, are becoming highly distributed compute platforms ideally suited to host emerging resource-intensive and latency-sensitive applications (e.g., industrial automation, extended reality, distributed AI). The end-to-end orchestration of such demanding applications, which involves function/data placement, flow routing, and joint communication/computation/storage resource allocation, requires new models and algorithms able to capture: (i) their disaggregated microservice-based architecture, (ii) their complex processing graph structures, including multiple-input multiple-output processing stages, and (iii) the opportunities for efficiently sharing and replicating data streams that may be useful for multiple functions and/or end users. To this end, we first identify the technical gaps in existing literature that prevent efficiently addressing the optimal orchestration of emerging applications described by information-aware directed acyclic graphs (DAGs). We then leverage the recently proposed Cloud Network Flow optimization framework and a novel functionally-equivalent DAG-to-Forest graph transformation procedure to design IDAGO (Information-Aware DAG Orchestration), a polynomial-time multi-criteria approximation algorithm for the optimal orchestration of NextG media services over NextG compute-integrated networks.

Figures (21)

• Figure 1: Illustration of a microservice-based XR application running over the device-edge-cloud continuum.
• Figure 2: Illustration of the lack of isomorphism between an information-aware service graph and its instantiation into the physical infrastructure: (a) depicts an information-aware service graph with colors indicating the information carried by each data stream; (b), (c), and (d) illustrate three different possible information-aware embedding into an 8-node network.
• Figure 3: Cloud-augmented graph, where green edges represent traditional network links indicating the availability of communication resources for transmitting information between nodes, and blue edges represent production, consumption, and computation capabilities at a given node.
• Figure 4: Example of a service graph, where edges represent data streams (commodities) and vertices service functions, for a NextG media application, in which streams from two sources go through tracking, synthesis, and rendering functions before being delivered to corresponding destinations.
• Figure 5: Illustration of how information aware overlapping of commodity flows results in replication of information flows. Commodities $k_2$ and $k_3$ carry the same information and are hence associated with the same object $o_2$. Hence, only one copy of the associated information flow needs to travel over common links $e_3$ and $e_4$.

Theorems & Definitions (42)

• Remark 1
• Remark 2
• Remark 3
• Remark 4
• Definition 1
• Remark 5
• Remark 6
• Remark 7
• Remark 8
• Definition 2