S ONS2014 Content Take Away
S
ONS2014Content Take Away
John Donovan, AT&T
How SDN enabled innovations will impact
AT&T’s plans to transform it’s infrastructure
Openness
Transformation & Innovation
Flexibility + Simplicity + SpeedNew Resources Model in vM
Redefining end point
Time required to provision services
SDN + NFVTelecom Network
Running on vM Simpler
Scalable
AT&T
www.at&tmorationspace.com
Build capabilities in AT&T Cloud for real time services
Mobile data traffic has increased 50000% over the last 7 years
Enhance business efficiency
Less CAPEX & OPEX
Transformation @ AT&TSimilar to Cloud
Computing
(Network Design)
Common Infrastructure,
Consumption model
Open Network principle >100 vendors collaboration
Domain 2.0 DODO
Budget for new architecture Ericson, Affirm Network - ecp, Metal Switch Core, Tel-ef – control software: open & flexible & work with existing devices
Open to new Ideas and maintain competitiveness
Doing with the WAN as the world has done to Data Centre
cloud centric workloads
Elasticity to Scale
Provision services in WAN
network
Open Process
Agile development
Domain 2.0 DODO
2014 Domain 1.5
Control on existing platforms – extend utilizing
New Platform in 2015 - NO overlay network – distributed cloud
2015 4600 Data Centre
microseconds of compromise that exist between middle and applications will be overcome by the milliseconds they avoid by
highly distributed, very fast fiber network
OSS
APIs – Do Policy & Provisioning
Software
Orchestrate in highly distributed network
Why running their services on your network?
Platform Extra move security into cloud system getter smarter as components get smarter through
communication
Privacy & Liability Guidelines Clarify boundaries – syntactic data build capability without
touching the raw 3 new technologies – control plane role, library, routing
system for data object – abstract syntactic data– no compromising customers
Radical reshaping entire white area
Early Road Map Items Excess part of network and core backbone = 2 highest
requirement and list amount of generated incremental benefit Data Centre – Latency tolerance (policy & authothenticy) = mobile
core converging wireless and wire line in universal services IMS
platform
Network Access Wireless – more like R&C pooling – really virtualization Complex (3G, 4G, wifi, AWS, WCS…) load very fast Optimized backbone switches and routing
Amin Vahdat, GoogleGoogle’s experience with Software Defined Network Function Virtualization at
Scale
• DE and Tech Lead for Networking, Google
• SAIC Professor of Computer Science and Engineering, UCSD
• ACM Fellow, Sloan Fellow, Duke University...
Networking @ GoogleNetwork Corporation Team
CLOUD
Infinite computer storage on demand
Fundamentally easier operational model (not there yet – true promise of the cloud – scale business)
Much higher of time – 3or4 nines of availability for services (multiplexing – leverage)
State-of-the-art infrastructure services – DoS, Load Balancing, Storage (can’t buy string out) – Offer these services = key to success
Programming models unavailable elsewhere – low lactency programming, massive IOPS
Andromeda Network Virtualization
Google Cloud Platform
CLOUD
SDN control of entire hardware/software – QoS, Latency, fault tolerance (Holistic Approach)
Virtualize SDN with NFV (Non-standard Network functionality per packet, can’t predict all the services, API for NFV = big opportunity)
Orchestrate & Manage: Network provisioning, High availability, Balanced virtual infrastructure (Network, Storage & Compute in right proportion for highest efficiency)
Andromeda Network Virtualization
Google Cloud Platform
Google Infrastructure
Google Global CDN (focus on driving down to milliseconds between Google services and end users
$2.9B Data Centre investments worldwide
Managing the energy in the cooling - efficiency 2-3x better
Google Infrastructure
Cluster Networking
- storage building level
- computation anytime anywhere
B4 SDN: Google Software Defined WAN
Google Infrastructure
First Google File System
Inspired HDFS
Inspired Big Data revolution
World largest NoSQL Implementation
Wide area consistence storage infrastructure for transaction across planet
New Challenges
Isolation
DDoS
Virtual IP networks
Network Function Virtualization
Mapping external services into internal namespace
Authentication, authorization, billing
Maintaining efficiency while doing all of the above
From Network Virtualization
EfficiencyBalance Compute, Storage
Capacity Bandwidth, Memory, Network
Bandwidth
How to provisioning your system:
Problem: Different application has different balanced point
SDN solution: SDN provisioning (Ser up the balance point of network compute and storage)v
EfficiencyNeed Fundamental
transformation in virtual networking
How to Spin up 1000 port virtual network with isolation, load balancing, external access, bandwdith provisionin?
How to deliver highest availability transparently to the end customer?
SDN + NFV
Provision an isolated, high-performance network across NIC, soft switch, storage, packet processor, fabric switches, Tor Audit correctness – transactional distributed operation (the
weakest link is whatever part you get wrong) Provision resources end to end QoS and availability
Logically centralized network management
Programmable packet processors for extensible network functionality
APIs for network application interoperability
SDN ROLE
SDN Opportunities and Challenges
“Middle boxes like functionality” SDN API for others to build their own network functionality with customized requirement
NFV
Andromeda Control Stack
Case study 1
Datapath logical view
Andromeda Network Datapath with intergrated programmable NFV Datapath pipelined, replicated multiple times as VM resources
scale out Critical optimization applied e2e (leveraging locality, forwarding
with less rules) Goal: near native performance, CPU efficiency
Case study 1
Network Performance
Case study 1
CPU Effiency
Case Study 2
Rapid provisioning of virtual networks
Fast ramp to peak load
Massive scale
Low Cost!
Network Virtualization @ Google
Delivering high performance shared computing infrastructure Enable new programming model Open up to customer to use same infrastructure
Logical centralized SDN control Orchestra across many different components
API for extensible NFV
Goals: native performance of hardware, full isolation, extensible NFV, high availability, scale out to many VMs.