Speaker: MOSHE GABEL, UNIVERSITY OF TORONTO
Topic: FROM CLOUD TO EDGE, AND BACK AGAIN: MAKING GEO-DISTRIBUTED COMPUTING EASY(ER)
Date: TUESDAY, APRIL 12, 2022
Time: 10:00 am – 11:00 am
Zoom: https://uwaterloo.zoom.us/j/97418382237?pwd=d21HcmtWK0ljQ2dSdXBVd2lOZkxEZz09
Meeting ID: 974 1838 2237
Passcode: 923325
Abstract:
As data becomes increasingly geo-distributed, we are hitting the limits of cloud computing. Scaling inside the cloud will not help if we cannot get data into the data center in the first place. Computing at the edge of the network, however, comes with new challenges: high-latency links with variable capacity, resource-constrained heterogeneous devices, and user mobility.
Unfortunately, our existing mechanisms for distributed computation are unsuitable for geo-distributed environments. Current system infrastructure is inefficient on the edge due to implicit assumptions from the cloud computing era that no longer hold, while current approaches for distributed data analysis are either too limiting or inaccessible to non-experts. Making the shift from cloud to edge requires a rethinking of how we design systems and algorithms.
In this talk I will present two projects that address this challenge. The first is Starlight, a complete redesign of the container deployment pipeline that dramatically accelerates software provisioning while maintaining backwards compatibility. On average, Starlight deploys and starts containers 3x faster than the industry-standard implementation, and 2x faster than the state-of- the-art approach. Next, I will present AutoMon: the first automatic approach for distributed approximation of arbitrary functions. Given source code that computes a function, such as the inference operation of a deep neural network, AutoMon automatically generates a communication-efficient scheme for approximating that function over the aggregate of distributed data streams.
Biography:
Moshe Gabel is a limited-term assistant professor in the Department of Computer Science at the University of Toronto, working on making geo-distributed data analysis more practical and accessible to typical software developers. He earned his Ph.D. in 2017 at the Technion - Israel Institute of Technology, having worked on communication-efficient algorithms. His work has appeared in top venues for data science (ICML, KDD) and systems (SIGMOD, NSDI, VLDB, ICDE), and has been recognized with two Best Paper Honorable Mentions. Finally, Moshe also works extensively on machine learning applications in pervasive health monitoring and in systems.
Website https://www.cs.toronto.edu/~mgabel/
All are welcome!