Event Title

On the hunt for a truly distributed global supercomputer

Session Number

Workshop Session 2

Location

Grainger Center for Inquiry and Innovation (B131)

Start Date

26-4-2018 11:45 AM

End Date

26-4-2018 12:30 PM

Abstract

Large scale computing is an important and growing part of modern life. It has applications in many areas including the rendering of videos, analysis of data about extraterrestrial life, design of proteins, discovery of cryptocurrency, simulation of physical or engineering systems, and a myriad of other applications. Most large scale computing platforms are limited to one general model. In this model, a single actor controls large amounts of computational resources. Users submit jobs to the resource, which then manages the actual allocation of computation to the job. Large scale services like Amazon or Google computing centers use this model.

In this workshop we will discuss this computational model and compare it to an alternative swarm-based model. A swarm may be thought of as a communicating group of agents whose capabilities together exceed those of the individual agents, even when acting individually in parallel. A swarm-based fully distributed computational model has many advantages over centralized models. Students will explore these advantages through an interactive activity.

Comments

Workshop includes student presentation:
Generating a Truly Global Swarm-Based Supercomputer
Presenters: Isaiah Crews, Andy Lennox, Lucien Putnam, and Vaishnavi Vanamala
Research Mentor: Dr. Sanza Kazadi, IMSA; Dr. Suranga Hettiarachchi, University of Indiana Southeast
Abstract:
Distributed computational platforms typically utilize a centralized model in which a central controller schedules and manages computations carried out on multiple independent processing units. This model is generally implemented in not only physical supercomputers but also in largely distributed computational networks. A second model is a fully distributed computational platform made up of potentially heterogeneous processors of vastly different capabilities in which computations can be initiated and scheduled by any participating processor. We propose the construction of this second distributed computational model. We will demonstrate, theoretically and in practice that such a computational platform can approach optimality and can enable adaptive distributed use across a variety of platforms and user needs. In the coming year, a prototype of Eximius will be completed and be used to test network optimization in the aspects of speed, and fault management and security. We will prototype Eximius according to results achieved through these tests.

Share

COinS
 
Apr 26th, 11:45 AM Apr 26th, 12:30 PM

On the hunt for a truly distributed global supercomputer

Grainger Center for Inquiry and Innovation (B131)

Large scale computing is an important and growing part of modern life. It has applications in many areas including the rendering of videos, analysis of data about extraterrestrial life, design of proteins, discovery of cryptocurrency, simulation of physical or engineering systems, and a myriad of other applications. Most large scale computing platforms are limited to one general model. In this model, a single actor controls large amounts of computational resources. Users submit jobs to the resource, which then manages the actual allocation of computation to the job. Large scale services like Amazon or Google computing centers use this model.

In this workshop we will discuss this computational model and compare it to an alternative swarm-based model. A swarm may be thought of as a communicating group of agents whose capabilities together exceed those of the individual agents, even when acting individually in parallel. A swarm-based fully distributed computational model has many advantages over centralized models. Students will explore these advantages through an interactive activity.