~/talk

Optimizing Network Provisioning through Cooperation

Talkat 19th USENIX Symposium on Networked Systems Design and Implementation (NSDI '22) · 2022-04-04..04-06· [networking, cloud-infrastructure]

speakers

name:Harsha Sharma

links:none on file

name:Parth Thakkar

links:none on file

name:Sagar Bharadwaj

links:none on file

name:Ranjita Bhagwan

links:none on file

name:Venkata N. Padmanabhan

links:none on file

name:Yogesh Bansal

links:none on file

name:Vijay Kumar

links:none on file

name:Kathleen Voelbel

links:none on file

resources

video:https://youtu.be/__gW8QfEbyQ

slides:none on file

summary

Cloud inter-DC provisioning uses application-declared deadlines and link failure time-series to cut capacity needs by about 30–45% on real substrate data. A two-step proof-of-concept pipeline first does rate-based provisioning for immediate demands, then a packing-style volume-based provisioning for deferrable demands; it clusters correlated links, samples time-to-repair/time-between-failure distributions, and simulates failures to schedule deferrable traffic into valleys. Results on a tier‑1 cloud and a large collaboration service show time‑shifting plus informed redundancy provisioning yields the reported savings.

AI-written summary · about

chapters (13)

related talks