Joint Compute-Caching-Communication Control for Online Data-Intensive Service Delivery

Yang Cai, Jaime Llorca, Antonia M. Tulino, Andreas F. MolischarXiv 2022 · DOI · arXiv

TL;DR

This paper introduces DI-DCNC, the first throughput-optimal control policy that jointly orchestrates computation, caching, and communication (3C) resources for data-intensive services like the Metaverse. By modeling complex service workflows as multi-pipeline flows on an "Augmented Layered Graph," the authors demonstrate significant gains in network capacity and resource efficiency compared to traditional, decoupled approaches.

What Problem Does This Paper Address?

Emerging "Metaverse" and Augmented Reality (AR) applications are incredibly demanding. They don't just require high bandwidth; they require the real-time fusion of live data (like a user's camera feed) with static digital assets (like 3D models stored in a database).

Current network control policies often handle these resources in silos: routing live streams is one problem, and caching static content is another. However, in a distributed edge cloud, these decisions are deeply coupled. Where you choose to process a video frame dictates where the static 3D assets must be sent, and vice versa. The authors tackle this "multi-pipeline flow control" challenge to prevent network bottlenecks and maximize the number of users a system can support simultaneously.

Methods

The researchers developed a comprehensive framework consisting of three main parts:

  1. Augmented Layered Graph (ALG): This is the "secret sauce." The authors transform the complex 3C problem into a specialized routing problem. The ALG creates virtual layers for each stage of a service's processing chain, allowing them to model live data pipelines and static asset pipelines as concurrent flows that must "merge" at specific processing nodes.
  2. DI-DCNC (Data-Intensive Dynamic Cloud Network Control): Using Lyapunov drift control theory, they designed an online policy that makes routing and processing decisions in real-time based on current queue lengths. It is mathematically proven to be "throughput-optimal," meaning it can stabilize the network for any demand within the theoretical maximum capacity.
  3. Joint 3C Orchestration: Beyond just moving data, the paper provides a Mixed Integer Linear Programming (MILP) formulation for optimal database placement (where to store the static assets) and two heuristic replacement policies (Rate-based and Score-based) to adapt to changing user preferences over time.

Datasets

The paper relies on numerical simulations rather than a specific public dataset. They modeled a grid-based Multi-access Edge Computing (MEC) network with 9 edge servers and one central cloud data center.

  • Traffic Model: User requests were modeled using i.i.d. Poisson processes.
  • Service Model: They simulated four different types of services, each with unique "scaling factors" (how data size changes after processing) and "merging ratios" (how many static objects are needed per live packet).
  • Popularity: For dynamic scenarios, they used a Zipf distribution to simulate shifting content popularity.

Key Insights and Ideas

  • The Multi-Pipeline Coupling: A key "aha moment" is the realization that processing location selection affects the transmission load of all input streams. Moving a computation closer to the live source might move it further from the database, potentially creating a worse bottleneck.
  • 3C Integration Expands the "Stability Region": By jointly optimizing, the network can support much higher arrival rates. The authors show that ignoring the transmission load of static objects (a common shortcut in older models) leads to significant performance degradation.
  • Asynchronous Updates: The "Score-Based" replacement policy suggests that you don't need to update every cache at once. Updating the single most "valuable" node (the one that reduces overall network weight the most) is an efficient way to adapt to new trends without overwhelming the backhaul network.

Key Results

  • Throughput Gains: DI-DCNC achieved a throughput of 1.05 Gbps, significantly outperforming benchmark algorithms like "Live-to-static" (920 Mbps) and "Static-to-live" (660 Mbps).
  • Resource Efficiency: In resource-constrained scenarios, the proposed joint policy saved roughly 50% of processing and transmission resources compared to baseline methods while maintaining the same delay requirements.
  • Optimal Placement: The MILP-based database placement was shown to be ~37% better than random placement, proving that where you cache is just as important as what you cache.

Limitations and Future Work

The authors acknowledge a few key areas for improvement:

  • Throughput vs. Delay: While DI-DCNC is throughput-optimal (it keeps the system from crashing), it is not necessarily delay-optimal. In low-congestion scenarios, it might choose longer paths that add unnecessary latency.
  • Scheduling Nuances: The current model treats live and static packets separately in the scheduling queues. A more advanced "incentive mechanism" that jointly schedules paired packets could further reduce latency.
  • Complexity: While the ALG simplifies the logic, the resulting MILP for database placement is still NP-hard, requiring approximation algorithms for very large-scale networks.

Future work is expected to focus on "lifetime-driven" decisions to minimize end-to-end latency for ultra-responsive Metaverse experiences.

Summary updated

Abstract

Emerging Metaverse applications, designed to deliver highly interactive and immersive experiences that seamlessly blend physical reality and digital virtuality, are accelerating the need for distributed compute platforms with unprecedented storage, computation, and communication requirements. To this end, the integrated evolution of next-generation networks (e.g., 5G and beyond) and distributed cloud technologies (e.g., fog and mobile edge computing), have emerged as a promising paradigm to address the interaction- and resource-intensive nature of Metaverse applications. In this paper, we focus on the design of control policies for the joint orchestration of compute, caching, and communication (3C) resources in next-generation distributed cloud networks for the efficient delivery of Metaverse applications that require the real-time aggregation, processing, and distribution of multiple live media streams and pre-stored digital assets. We describe Metaverse applications via directed acyclic graphs able to model the combination of real-time stream-processing and content distribution pipelines. We design the first throughput-optimal control policy that coordinates joint decisions around (i) routing paths and processing locations for live data streams, together with (ii) cache selection and distribution paths for associated data objects. We then extend the proposed solution to include a max-throughput database placement policy and two efficient replacement policies. In addition, we characterize the network stability regions for all studied scenarios. Numerical results demonstrate the superior performance obtained via the novel multi-pipeline flow control and 3C resource orchestration mechanisms of the proposed policy, compared with state-of-the-art algorithms that lack full 3C integrated control.
BibTeX
@misc{cai2022joint,
  author = {Yang Cai and Jaime Llorca and Antonia M. Tulino and Andreas F. Molisch},
  title = {{Joint Compute-Caching-Communication Control for Online Data-Intensive Service Delivery}},
  year = {2022},
  eprint = {2205.01944},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2205.01944},
}
Figure