Multi Commodity Flow Problem in Programmable Networks

When:
Monday, December 14, 2020, 9:00 am - 10:00 am PSTiCal
Where:
Virtual
Type:
Institute Talk hosted by the Networking and Cybersecurity Division
Speaker:
Ganesh C Sankaran, UC Davis
Video Recording:
https://usc.zoom.us/j/99230009536
Description:

Abstract:

Network switches primarily perform routing and scheduling of packets. Recently, programmable networks have become a reality. These network switches are organized as pipeline stages with each stage providing specific quantum of lookup and ALU resources. This enables custom functions outside the ambit of network functions to be deployed on network switches. This has opened up a new area called in-network computing. Couple of motivational usecases are presented for this new paradigm. Multiple functions competing for the lookup and ALU resources is similar to the multi-commodity flow problem. Here, the key bottleneck is the quantum of resources available for custom function implementations.

This talk first introduces programmable networks and its building blocks. Then, it introduces in-network computing and its relation to multi-commodity flow problem. Considering N distinct computing tasks, the total resources required is O(N). In other words, the quantum of resources available for a task shrinks by a factor of N. We then show how traditional computing paradigms suffer from this limitation.

Then, we propose an active packet paradigm. This paradigm requires minimum footprint on network switches for task processing. We show that the amount of resources required for this approach is O(1). Specifically, it is independent of the number of distinct tasks (or N).

Finally, we conclude by showing how compute servers can leverage this new paradigm in a seamless manner.

Speaker Bio:

Ganesh C Sankaran is a post-doctorate scholar in the Computer Science department at University of California at Davis. His research interests include programmable networks and distributed computing, particularly in the context of realizing higher-layer functions on top of programmable networks. He received his Ph.D. in Computer Science and Engineering from the Indian Institute of Technology Madras. He has rich industy experience working for Cisco Systems and Dell Networking.

ISI Host:  Terry Benzel, Division Director - Networking and Cybersecurity

« Return to Upcoming Events