Dynamo: A Framework for Dynamic Real-Time Workloads in a Reliable Cloud

Abstract

Short work statementThis research project proposes to develop Dynamo: A Framework for Dynamic Real-Time Workloads in a Reliable Cloud. The proposed 3-year effort will focus on the development and characterization of heterogeneous two-type (~big.Little~) real-time scheduling and cloud-level continuity of operations-as-a-service. The Dynamo project will leverage the existing resource management framework developed under the ANCHORS project and will extend the framework to accommodate heterogeneous real-time scheduling and fault tolerance.ObjectiveThis research project proposes to develop Dynamo: A Framework for Dynamic Real-Time Workloads in a Reliable Cloud. Dynamo builds on the successful ANCHORS real-time cloud framework that enabled high performance realtime workloads within the OpenStack cloud platform.ApproachToday~s cloud frameworks largely assume a homogeneous, IT-centric design, mostly suitable for traditional enterprise workloads. Mission-centric clouds, on the other hand, are composed of a heterogeneous mix of traditional and realtime workloads, general-purpose and heterogeneous architectures, and high performance and embedded resources.The Dynamo team will conduct research covering a range of heterogeneous real-time processing and scheduling techniques with an emphasis on reliability and continuity of operations support for real-time workloads. This research project proposes Dynamo: A Framework for Dynamic Real-Time Workloads in a Reliable Cloud. Dynamo builds on the successful ANCHORS real-time cloud framework that enabled high performance real-time workloads within the OpenStack cloud platform. Dynamo will innovate in the following ways:~ Heterogeneous real-time workloads: The Dynamo project will develop novel scheduling strategies for virtualized twolevel heterogeneous architectures. This will enable the first online scheduling strategy for two-level heterogeneity, and the first to support dynamically adjusting the heterogeneity by way of a real-time hypervisor. This support is intended to integrate directly into the existing ANCHORS framework, enabling additional classes of heterogeneous real-timeworkloads by dynamically manipulating resource allocation on a per-core or per-virtual CPU basis.~ Continuity of Operations ~ COOP-as-a-Service: The Dynamo project will develop a framework to enable Continuity-Of-Operations-as-a-Service that reflects the unique requirements of DOD mission applications and their deployment to the cloud. The Dynamo project will support heterogeneity in both applications and techniques for continuity of operations, with particular focus on real-time applications.~ Customizable reliability: Today~s ~cloud model~ pushes the primary responsibility of reliability onto the end-user,typically by way of horizontal scaling and/or restrictive Platform-as-a-Service models. While these PaaS strategies are helpful, particularly for those applications that map cleanly to technologies like MapReduce or Spark, many applications are poorly suited for these types of programming models. Real-time command and control applications, proprietary and legacy software, and non-data-centric real-time applications are especially unsuitable. Dynamo willdevelop a suite of reliability strategies based on continuity of operations requirements, and will develop a runtime monitor capable of enforcing an application~s specific reliability requirements.Overall merits & ONR mission relevanceComputing resources on Navy platforms are likely to be tasked with a wide variety of computing applications andscheduling may be more challenging because of the need to share these more centralized processing resources across a larger user and application base. For example, on a ship, a common set of computing resources may be required to support beamsteering for a phased array radar, anti-jamming algorithms, automated target recognition, and ship control functions simultaneously. This technology, however, requires a highly parallel and hi

Document Details

Document Type
DoD Grant Award
Publication Date
Sep 30, 2016
Source ID
N000141612887

Entities

People

  • John Paul Walters

Organizations

  • Office of Naval Research
  • United States Navy
  • University of Southern California

Tags

Fields of Study

  • Computer science
  • Engineering

Readers

  • Distributed Systems and Data Platform Development
  • Parallel and Distributed Computing.
  • Space/Atmospheric Physics.

Technology Areas

  • Fully Networked C3
  • Fully Networked C3 - Command and Control