[Colloquium] Rohan Kumar MS Presentation/July 16, 2020

[Jessica Garza]

This is an announcement of Rohan Kumar's MS Presentation. Rohan is a student in the Bx/MS program.

Here is the Zoom link to participate: 
https://uchicago.zoom.us/j/96030791303?pwd=ZjZXSjBsRGlmcTVyblNpZElGK2tDUT09 <https://urldefense.com/v3/__https://uchicago.zoom.us/j/96030791303?pwd=ZjZXSjBsRGlmcTVyblNpZElGK2tDUT09__;!!BpyFHLRN4TMTrA!u4JO_qv7aODU2ciogaEnnA_Aff1r-__NJyJOwG8S383OWyKjYfBkJY0gkeT3HFhA6d0$>

----------------------------------------------------------------------------------------------------------------------------------------

Date: Thursday, July 16, 2020

Time: 11 AM, CST

Location: remote via Zoom

M.S. Candidate: Rohan Kumar

M.S. Paper Title: Coding the Computing Continuum: Enabling Fluid Execution of Functions in Heterogeneous Computing Environments

Abstract:

 With network speeds faster than they have ever been before, the once prominent dis- tinction between local and remote execution is now rapidly disappearing. This has allowed users to begin exploring methods of running their analyses on remote devices which are not only more powerful, but often more suited to their analyses. No longer are users constrained by the limited capabilities of their local devices. Instead, with little effort, they can now access high-performance clouds and clusters, where their execution needs can be more than satisfied. As such, the world is experiencing a shift away from general-purpose devices that can be used for diverse applications. The modern age of computing is one of specialization, with special-purpose hardware being carefully optimized for particular applications. In such an environment of ubiquitous heterogeneity, there is an increasing need to automate the selection of hardware for arbitrary computations. However, the question of where to run a computation is full of complexities, and depends on a multidimensional array of factors, including task performance, communication costs, and resource-provisioning delays. The goal of this work is to develop techniques to model these different factors as part of a single comprehensive framework for heterogeneous execution. To accomplish this goal, we build upon existing work on function-serving platforms, since functions represent a convenient ab- straction for remote computation. Our contributions in this work are threefold. First, we demonstrate that there are inescapable trade-offs between compute times, network speeds, start-up delays, and other latencies, which must be accounted for in the choice of hardware and location for computations. Second, we present Delta, a framework for heterogeneous function execution that is designed to account for these trade-offs, and provides improved performance by taking advantage of the heterogeneity in compute resources. Finally, we de- fine a research agenda in this space, and motivate future work with the intention of building a world where fluid computation can exist.

Advisors: Ian Foster and Kyle Chard

Committee Members: Ian Foster, Kyle Chard, Raul Castro Fernandez




Jessica Garza
Assistant Director of Undergraduate Studies
Department of Computer Science
The University of Chicago
Covid-19 Resources <https://cs.uchicago.edu/remote2020/>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20200716/32e4b849/attachment-0002.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Rohan Kumar - MS Thesis.pdf
Type: application/pdf
Size: 5679255 bytes
Desc: not available
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20200716/32e4b849/attachment-0001.pdf>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20200716/32e4b849/attachment-0003.html>