SIFT: Scheduling Interactive Flows to Tame Performance Tails

Abstract

Future tactical grids for the Navy/Marine Corps are becoming increasingly sophisticated and heterogeneous. These consist of distributed systems involving compute infrastructure, devices, war?ghting groups, and eets which interact with each other on a network. Different network entities must work in concert to provide timely, seamless, and robust connectivity under dynamic (and even hostile) environments and to achieve mission-critical objectives - such as minimizing various performance tails - going beyond traditional considerationslike throughput and congestion. Traditional techniques that are agnostic to suchrequirements do not perform well in optimizing these collections, because network management has largely been framed as an optimization of individual ow-level metrics. We propose SIFT, a novel framework for Scheduling Interactive Flows to Tame performance tails in naval tactical grids. This framework abstracts the ow interactions/dependencies in the network and enables e?cient algorithmic solutions for optimizing network performance tails. To the best of our knowledge, there is no single model that systematically captures large variety of ow interactions/dependencies in prevalent networking paradigms. SIFT aims to provide the ?rst systematic optimization framework that uni?es various types of ow interactions and interdependence - including Combinatorial-success,Fate-sharing, Priority-control, denoted as CFP model - that are observed (and often overlooked) in many aspects of mission-critical maritime operations. A key in the scheduling paradigm is the uncertainty in network ows, congestion, background tra?c, unknown future behavior, etc., which will be studied using di -erent metrics tailored to uncertainty (e.g., performance tail) and online algorithm with guarantees. Relevance to ONR: ONR has a keen interest in developing an operational architecture for distributed maritime operations, connecting distributed units into groups and distributed groups into eets. At the core of this architecture is the tactical grid, in which di -erentnetwork entities must work in concert to provide timely, seamless, and robust connectivity under dynamic (and even hostile) environments and to achieve mission-critical objectives (such as minimizing performance tails) going beyond traditional considerations. This proposal seeks to explore e?cient and resilient optimization algorithms that address tra?cinterdependence in future maritime operations. If successful, it can provide a transformative optimization and planning engine for the US Navy s network.

Document Details

Document Type
DoD Grant Award
Publication Date
Feb 17, 2020
Source ID
N000142012146

Entities

People

  • Tian Lan

Organizations

  • George Washington University
  • Office of Naval Research
  • United States Navy

Tags

Fields of Study

  • Computer science

Readers

  • Computer Networking
  • Distributed Systems and Data Platform Development