SDN for the Cloud

• SDN 41
• cloud 1

• slide
  • Consistent cloud design principles for SDN
  • Physical and Virtual networks, NFV
    • Hyperscale SDN: All Policy is in the Host
    • Regional Network Controller Stats
    • Azure SLB: Scaling Virtual Network Functions
  • Integration of enterprise & cloud, physical & virtual
• Keynote: SDN for the Cloud

paper, slide

slide

WAN

Motivating scenario network engineering

Innovation: infer traffic control routing, centralize control to meet network-wide goals

WAN

TomoGravity ✓ ACM Sigmetrics 2013 Test of Time Award

RCP Usenix NSDI 2015 Test of Time Award

4D ACM Sigcomm 2015 Test of Time Award

Cloud

VL2, sigcomm 2009

scale-out L3 fabric, SDN and NFV

per customer virtual networks

Cloud provided the killer scenario for SDN

Virtualized Data Center for each customer — prior art fell short

High scale fault tolerant distributed systems and data management

VL2 → Azure

Consistent cloud design principles for SDN

Scale-out N-Active Data Plane

Embrace and Isolate failures

Centralized control plane: drive network to target state

Resource managers service requests, while meeting system wide objectives Controllers drive each component relentlessly to the target state Stateless agents plumb the policies dictated by the controllers

Challenges of Scale

number of paths, dynamic set of gray failure modes, chasing app latency

Azure State Management

statesman

App I: Automatic Failure Mitigation

App II: Traffic Engineering Towards High Utilization (SWAN)

• Azure Scale Monitoring — Pingmesh
• Azure Cloud Switch — Open Way to Build Switch OS

Azure Cloud Switch OS to manage the switches as we do servers

Physical and Virtual networks, NFV

• Azure is the hub of your enterprise, reach to branch offices via VPN
• VNet is the right abstraction, the counterpart of the VM for compute
• Efficient and scalable communication within and across VNets

Hyperscale SDN: All Policy is in the Host

Challenges for Hyperscale SDN Controllers - scale up to 500k+ Hosts in a region - scale down to small deployments too - millions of updates per day - support frequent updates without downtime

Regional Network Controller Stats

API execution time

• Read : < 50 milliseconds
• Write : < 150 milliseconds

Varying deployment footprint

• Smallest : <10 Hosts
• Largest : >100 Hosts

Azure SLB: Scaling Virtual Network Functions

Data Center-Scale Distributed Router

extrem scale: 10k customers, 10k routes each → 100M routes

Integration of enterprise & cloud, physical & virtual

Virtual Filtering Platform (VFP)

acts as a virtual switch inside Hyper-V VMSwitch

Flow Tables

a typed Match- Action-Table API to the controller

the Right Abstraction for the Host

http://research.microsoft.com/en-us/um/cambridge/events/sigcomm2015/papers.html

Keynote: SDN for the Cloud

ABSTRACT Cloud computing is a new paradigm, which touches everything from enterprise data centers, to wide area networks, to massive scale data centers, with millions of servers. Coordinating such large systems requires a new way to control the network — setting a target state for each component and driving each to the goal state, so that the network meets global objectives, in a sea of growth and change: customers creating and adapting virtual networks at amazing pace, while new features are rolled out, new data centers are built out and older ones decommissioned. Software Defined Networking (SDN) is a key enabling technology, built on principles of delivering direct control and network-wide views. At Microsoft Azure, we pioneered these principles, and built them into the fabric of the Cloud.

cloud optimized, low cost, high performance, super reliable and automated

products

Virtual Layer-2 (VL2), Virtual Networks (VNets), Load Balancing (Ananta), Data Center TCP (DCTCP)