1 / 9

SDWAN: Inter-Datacenter Transport Scheduling

This topic explores software-defined wide area network (SDWAN) and its role in scheduling inter-datacenter transport. It covers challenges faced in DC transport, DC TCP, and Fastpass and discusses solution spaces such as CSFQ, D3, PDQ, pFabric, WFQ, DeTail, GPS, RED, ECN, XCP, RCP, DCTCP, D2TCP, and FCP.

deforest
Download Presentation

SDWAN: Inter-Datacenter Transport Scheduling

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. CS434/534: Topics in Network SystemsInter-datacenter Transport Scheduling: Software-Defined WAN (SDWAN)Yang (Richard) YangComputer Science DepartmentYale University208A WatsonEmail: yry@cs.yale.eduhttp://zoo.cs.yale.edu/classes/cs434/

  2. Road Map • Cloud data center (CDC) applications/services • Fine-grained dataflow programming (e.g., Web apps) • Coarse-grained dataflow (e.g., data analytics) • Distributed machine learning using parameter server • DC scheduling • DC cluster resource scheduling • DC transport scheduling • Overview • DCTCP • Fastpass • WAN scheduling • CoFlow scheduling

  3. Admin • Instructor office hours • Tuesday: 1:30-2:30 • Thursday: 3:00-4:00 • Fridays: 1:30-2:30 pm

  4. Admin • Projects • Milestones (exactly 4 weeks left) • 4/17 (T+1 week): Finished reading major related work; a google doc listing related papers (at least 4 papers) • 4/24 (T+2 weeks): Finished architecture design ( slides/write up of architecture, including all key components) • 5/1 (T+3 weeks): Initial, preliminary evaluations (slides/write up, about experiment/analysis setup) • 5/8 (T+4 weeks) 5:30 pm, final report due

  5. Admin • Current on-going student projects • Automatic, High-level SDN Compiler using Flow Algebra • SDN Update Algebra with Resource, Ordering Constraints • Multi Control Plane Switching using Distributed Verification in MANET using OLSv2 • Toward High-Level Programming for 5G using Integrated SDN/NFV • Design and Implementation of Distributed, Consistent, Fast Data Store • Distributed Robots as Mobile Datacenters for Shared Learning • Scaling Automatic, Consistent Web App Programming using Noria • Distributed, Federated Learning through Autoencoder • Scaling Distributed Machine Learning for Geo-Distributed Data Analytics • Scaling Fastpass to PB using Partition and Aggregation • Distributed, Encrypted Data Store for Global, Community Data

  6. Recap: DC Transport Challenges • Challenging setting: • Mixed elephants and mice traffic with potential high burst traffic • Challenging requirement: • Achieving high throughput for elephants and low latency for mice, despite potential high burst

  7. Recap: DC TCP • Problem of TCP: large, full buffer to absorb drastic rate cut to achieve high tput, but result in long latency and bad tolerance of burst traffic • Basic ideas • Switch using ECN: Mark packets when Queue Length > K (how big should K be?). • Host: Maintain running average of fraction of packets marked (α) and adjust using α

  8. Recap: Fastpass • Goals: low latency, high tput, multiple objectives using centralized scheduling • Key technical subproblems to solve • rate allocation (how?) • scaling rate allocation (multi-core, pipelining) • path selection (how?) • scaling path selection (efficient matching alg using Euler paths)

  9. Recap: Solution Space Covered So Far Flow: Transfer of data from a source to a destination CSFQ D3 PDQ pFabric WFQ DeTail GPS RED ECN XCP RCP DCTCP D2TCP FCP 1980s 1990s 2000s 2005 2010 2015 Per-Flow Fairness Flow Completion Time Individual, independent flows do not reflect communication behaviors common in data-parallel applications

More Related