Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LOAM: Low-latency Communication, Caching, and Computation Placement in Data-Intensive Computing Networks

Jinkun Zhang, Edmund Yeh

Abstract

Deploying data- and computation-intensive applications such as large-scale AI into heterogeneous dispersed computing networks can significantly enhance application performance by mitigating bottlenecks caused by limited network resources, including bandwidth, storage, and computing power. However, current resource allocation methods in dispersed computing do not provide a comprehensive solution that considers arbitrary topology, elastic resource amount, reuse of computation results, and nonlinear congestion-dependent optimization objectives. In this paper, we propose LOAM, a low-latency joint communication, caching, and computation placement framework with a rigorous analytical foundation that incorporates the above aspects. We tackle the NP-hard aggregated cost minimization problem with two methods: an offline method with a 1/2 approximation and an online adaptive method with a bounded gap from the optimum. Through extensive simulation, the proposed framework outperforms multiple baselines in both synthesis and real-world network scenarios.

LOAM: Low-latency Communication, Caching, and Computation Placement in Data-Intensive Computing Networks

Abstract

Deploying data- and computation-intensive applications such as large-scale AI into heterogeneous dispersed computing networks can significantly enhance application performance by mitigating bottlenecks caused by limited network resources, including bandwidth, storage, and computing power. However, current resource allocation methods in dispersed computing do not provide a comprehensive solution that considers arbitrary topology, elastic resource amount, reuse of computation results, and nonlinear congestion-dependent optimization objectives. In this paper, we propose LOAM, a low-latency joint communication, caching, and computation placement framework with a rigorous analytical foundation that incorporates the above aspects. We tackle the NP-hard aggregated cost minimization problem with two methods: an offline method with a 1/2 approximation and an online adaptive method with a bounded gap from the optimum. Through extensive simulation, the proposed framework outperforms multiple baselines in both synthesis and real-world network scenarios.
Paper Structure (20 sections, 8 theorems, 96 equations, 8 figures, 2 tables, 2 algorithms)

This paper contains 20 sections, 8 theorems, 96 equations, 8 figures, 2 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Network model
  3. Cache-enabled computing network
  4. Node-based strategies
  5. Problem formulation
  6. Offline method: $1/2$ approximation
  7. A "submodular + concave" reformulation
  8. Algorithm with $1/2$ approximation
  9. Online adaptive solution
  10. Closed-form marginals
  11. KKT condition and modification
  12. Interpretation and corollaries
  13. Online adaptive algorithm
  14. Simulation
  15. Conclusion
  16. ...and 5 more sections

Key Result

Lemma 1

Problem Objective_gain is a "submodular + concave" maximization problem. Specifically, $M(\boldsymbol{\phi})$ is non-negative monotonic DR-submodularDR-submodular function is a continuous generalization of submodular functions with diminishing return. See, e.g., bian2017guaranteed, for more informat

Figures (8)

  • Figure 1: (a) Could computing, server stores all data and performs all computations. (b) Offloading computations, retrieving data from server WiOpt22. (c) Offloading computations and caching data objects kamran2021deco. (d) Reusing computation results.
  • Figure 2: Flow level behavior of nodes $j \to i \to l$. We only mark the flows of CI and DI. Flows of CR/DR are on the same path as CI/DI, in the reversed direction.
  • Figure 2: Network Scenarios
  • Figure 3: Sample network satisfying Corollary \ref{['cor_cache_1']}. Single-layered caches and computing nodes are equipped near users. Requests are routed to servers if not fulfilled at the caches.
  • Figure 4: Normalized total cost $T$ of different methods in multiple network scenarios
  • ...and 3 more figures

Theorems & Definitions (8)

  • Lemma 1
  • Theorem 1: Theorem 3.10 mitra2021submodular
  • Lemma 2
  • Theorem 2
  • Corollary 1
  • Corollary 2
  • Corollary 3
  • Theorem 3