Back To Schedule
Tuesday, October 2 • 10:25 - 10:50
Things We Can Learn from Simulating DC Networks

Sign up or log in to save this to your schedule, view media, leave feedback and see who's attending!

Data Center networks carry the superposition of a large number of workloads resulting from complex communication patterns between many tiers of services and applications. It is particularly interesting to model a background traffic pattern based on a subset of the existing traffic, and impose a foreground workload based on the most demanding of the emerging traffic demands. For the foreground traffic, we propose to model a traffic pattern that anticipates disaggregation trends, and storage is a reasonably well understood and relevant instance of disaggregation.In our context disaggregation means that a resource that was traditionally deployed as part of a server I/O subsystem is now remote and accessed over the Data Center network. We do simulate these two superimposed workloads at reasonable scale, with TCP endpoints included, and with fairly accurate queuing models for the Data Center switches. We find that results sometimes match the intuition, and sometimes do not.


Mark Griswold

Distinguished Engineer, Broadcom

Ariel Hendel

Infrastructure Technologist, Facebook

Tuesday October 2, 2018 10:25 - 10:50 CEST