Transitioning to an Automated, Software Driven
Cloud Network
David Murray - Systems Engineer EMEA - dm@arista.com
Enterprise DC"
1990+ 
Virtualized DC "
2000+
Private/Public Cloud
2010+
Hybrid Cloud "
Radar
Internet
Applications Architectures are Evolving
Enterprise DC"
1990+ 
Virtualized DC "
2000+
Private/Public Cloud
2010+
Hybrid Cloud "
Radar
Internet
Applications Architectures are Evolving
Solution - More Network Complexity?
Simplify
Build on Simple Foundations
• Vanilla IP networks
• Leaf-Spine topologies
• Network level redundancy
• Repeatable elastic units
• Robust APIs
Source: facebook.com
MLAG / LACP BGP / OSPF + ECMPSpline™ / LACP
VXLAN: Network
Virtualization



 100 to 10,000
 500 to 100,000+
 500 to 100,000+
Host Scale: 0 to 2,000
100,000+
 500,000+
 1,000,000+
VM Scale: 20,000+
Open standards meet all scaling requirements
Automate
Why automate (the bad)?
	
  $84,000.00	
  	
  
	
  $168,000.00	
  	
  
	
  $252,000.00	
  	
  
	
  $336,000.00	
  	
  
	
  $420,000.00	
  	
  
	
  $504,000.00	
  	
  
	
  $588,000.00	
  	
  
	
  $672,000.00	
  	
  
15	
   30	
   45	
   60	
   75	
   90	
   105	
   120	
  
Cost of an outage
Minutes
•  59% of F500s experience ≥1.6hrs downtime per week 
–  (Dunn & Bradstreet)
•  Average outage lasts 200 minutes 
–  (IT Process Institute)
•  Downtime average 87hours/year 
–  (Gartner)
•  Downtime costs $5600/minute 
–  (Ponemon Institute)
•  49% of outages are caused by human error
–  (IEEE)
80% of IT activity, time and dollars are spent
keeping the lights on
- http://www.gartner.com/newsroom/id/497088
Key wins for automation:
•  Time to delivery/repair
–  Zero Touch Provisioning and Replacement of
hardware
•  Design stability
–  Config validation and application
•  Avoiding grey failures
–  Automated upgrades
•  Pro-active fault management
–  Remediation and fault isolation
Source: facebook / nanog.org
Complexity
Do-It-Yourself
Custom development for
integration
Software development
resources & approach
DevOps
Leveraging existing tools
(Puppet, Chef etc)
Integration & customization
via scripts
Turnkey
Limited solutions today, focus on
legacy models
Customers need turnkey solution
for cloud-automation!
Approaches to Network Automation
Cloud Titan
High Tech
Enterprise / SP
Traditional
Enterprise / SP
10,000
network elements per engineer
1,000
network elements per engineer
100
network elements per engineer
© 2015 Arista - Confidential
Low Inter-Process Communication
Arista EOS Architecture
Efficient Publish/Subscribe
Publish
Notify
 PIM
SNMP
 BGP
Driver
 SysDB
MLAG
STP
 eAPI
IGMP
Linear Cloud Scale
Arista EOS: Resilient Cloud Scale Software
© 2015 Arista - Confidential
Open &"
Programmable
Merchant Silicon
SysDB
CLI
SNMP
VXLAN"
VM"
Tracer
OSPF
STP
ASIC"
Driver
SDN"
APIs
LAG
BGP
Modular &"
Resilient
Universal Cloud"
Network
Customer Driven"
Flexibility
Cloud Scale"
Economics
EOS
vEOS
EOSPrinciples
§  EOS Network Wide Services
§  Single network control point for:
•  Third party controllers
•  Orchestration services. 
§  Turnkey workflow automation
CloudVision
Workflow Automation Services
Workload Orchestration Services
EOS Network Wide Services
Open APIs & Partner Access
Arista’s Universal Cloud Network
Arista EOS CloudVision
© 2015 Arista - Confidential
EOS Devices
Workflow Automation
Services
CloudVision Framework
Management, Monitoring
Tools
Orchestrators and Overlay
Controllers
Open APIs
Open RESTful APIs
EOS Core Services
Workload "
Orchestration Services
CloudVision
eXchange (CVX)
OVSDB, eAPI, Plugins
3rd Party
Integration
CloudVision Portal
(CVP)
© 2015 Arista - Confidential
CloudVision Exchange
Platform for Automation and Visibility across the Network
Network-Wide Database
Aggregation of Network
wide ‘Sysdb’
Abstraction of the "
physical network
Single integration point to
the network
More scalable controller
integration
Realtime data collection
and Historic trending
State-sync
Cloud
Orchestrators
Overlay
Controllers
www.arista.com
Network
Services
Network"
Layer
Network Control
Point
OVSDB
JSON
Points of "
Integration
Single point of
integration to the
physical infrastructure
Web-based GUI
 CLI
Platform for Automation and Visibility across the Network
Network-Wide Database
State-sync
© 2015 Arista - Confidential
Overlay "
Controller
Scaling Controller Integration
18
OVSDB
Overlay "
Controller
Network Layer
Controller Layer
10x
Improvement
OVSDB
 SysDB 
State Sync
Topology/Device
Dependent
Topology/Device
Abstraction
Traditional
Approach
CloudVision
Approach
© 2015 Arista - Confidential
CloudVision Portal
CloudVision Portal Key Features 
20
Network Rollback
Efficiently revert the network
back to a known good state 
Overlay/Underlay Visibility
Physical and logical layout, VNI
mapping and performance
hotspots
Provisioning
Turnkey solution for
automating network
operations
Change Management
Snapshot of pre- and post-
changes for better control
Network-wide Upgrade
Automate software upgrades
across groups of devices
Maintenance
Mode
Production
Mode
Single Pane of Glass
Point and click interface to
simplify bulk tasks
© 2015 Arista - Confidential
Time t
Time t-1
✔
 ✗
CVP - A Single Interface to Automation
© 2015 Arista - Confidential
CloudVision Core Services
Manual box-by-box
operations. Difficult to
correlate network-wide.
Network-wide
synchronization for
topology level visibility.
No ability to monitor
code versions and alert
on software bugs
Services Dashboards,
with proactive
notification tools
Separate vendor tools
for Inventory, Syslog,
Management,
Compliance & other
infrastructure reporting
Consolidated reporting
view within a single
application
Typically pair of non-
customizable NMS tools
built for limited sizing,
scale & deployments
Big Data Infrastructure
fully customizable web
GUI
Primary
 Secondary
Initial
Node1
Object X
(Active)
Object Y
(Active)
Object Z
(Backup)
Node3
Object Y
(Backup)
Object B
(Active)
Node2
Object X
(Backup)
Object Z
(Active)
Object B
(Backup)
Node1 Failure
Node3Node2
Offline /
Failure
Node1
Object X
(Active)
Object Y
(Active)
Object Z
(Backup)
Object Y
(Backup)
Object B
(Active)
Object X
(Backup)
Object Z
(Active)
Object B
(Backup)
Node1 Node2 Node3
Node1 Node2 Node3 Node4 Node5
Initial
Add 2 Nodes
Before 
 With CloudVision
© 2015 Arista - Confidential
Cloud-like automation for everyone
Turnkey Provisioning
§  Automatically provision switches centrally
from a network wide view
§  Single tool & view to manage code,
configuration, scripts etc.
§  Manage switch throughout switch’s DC
life cycle using single provisioning
process
§  Full configuration coverage, including
VXLAN
§  Configuration compliance validation
© 2015 Arista - Confidential
Scripting
Image
Configuration
Configurations
Images
Scripts
ZTP/ZTR
Existing
Approaches
With Arista EOS
CloudVision
Change Management
§  Perform graceful change controls
without impacting applications
§  Automatic snapshots pre & post change
with delta reporting in network wide view
§  Useful for Smart System Upgrades
(SSU) or any change control
§  Integration into Workflow Tools, such as
Service Now
© 2015 Arista - Confidential
Confidence for Change Controls
go-
forward
Productio
n
Maintenanc
e
Roll-
back
Existing
Approaches
With Arista EOS
CloudVision
Configurations
Images
Scripts
✔
✔
 ✗
 ✗
Availability: Network-wide Rollback
§  Network-wide scope: Images,
configuration, scripts
§  Historical state used to recover to
previous network state.
§  Rolling snapshots during change
controls making it easier to roll-back
during issues
§  Reduce the time-to-recovery during
maintenance windows
Safety tool for maintenance window recovery
A time machine for 
your network

© 2015 Arista - Confidential
CloudVision: Platform for Partners
Partner for NSX Controller for
overlay networking  VROps for
visibility
Partner for Converged System
Reference architecture,
including HP OneView
Partner for Controller integration
for Transport networks
Partner for OpenStack
integration
Partner for L4-7 services
integration
Partner for Converged systems
Partner for L4-7 services
integration
Partner for OpenStack
integration
Partner for Solution integration,
including ASM
Partner for Windows Server
2016 integration for network
controllers
Partner for Controller integration
for overlay networks
© 2015 Arista - Confidential
Fast path to Cloud: Turnkey solution for Automation and Orchestration
EOS is now a “Network Wide Service” 
Opex Reduction: By enabling automation of simple cloud services 
Extends Arista’s EOS Platforms for Software Defined Cloud Networking
1
2
3
4
CloudVision Summary
© 2015 Arista - Confidential
Thank You - Any Questions?
dm@arista.com

PLNOG15: Arista EOS Cloud Vision: Pivotal point in workload orchestration and workflow automation - David Murray

  • 1.
    Transitioning to anAutomated, Software Driven Cloud Network David Murray - Systems Engineer EMEA - dm@arista.com
  • 2.
    Enterprise DC" 1990+ VirtualizedDC " 2000+ Private/Public Cloud 2010+ Hybrid Cloud " Radar Internet Applications Architectures are Evolving
  • 3.
    Enterprise DC" 1990+ VirtualizedDC " 2000+ Private/Public Cloud 2010+ Hybrid Cloud " Radar Internet Applications Architectures are Evolving Solution - More Network Complexity?
  • 4.
  • 5.
    Build on SimpleFoundations • Vanilla IP networks • Leaf-Spine topologies • Network level redundancy • Repeatable elastic units • Robust APIs Source: facebook.com
  • 6.
    MLAG / LACPBGP / OSPF + ECMPSpline™ / LACP VXLAN: Network Virtualization 
 100 to 10,000 500 to 100,000+ 500 to 100,000+ Host Scale: 0 to 2,000 100,000+ 500,000+ 1,000,000+ VM Scale: 20,000+ Open standards meet all scaling requirements
  • 7.
  • 8.
    Why automate (thebad)?  $84,000.00      $168,000.00      $252,000.00      $336,000.00      $420,000.00      $504,000.00      $588,000.00      $672,000.00     15   30   45   60   75   90   105   120   Cost of an outage Minutes •  59% of F500s experience ≥1.6hrs downtime per week –  (Dunn & Bradstreet) •  Average outage lasts 200 minutes –  (IT Process Institute) •  Downtime average 87hours/year –  (Gartner) •  Downtime costs $5600/minute –  (Ponemon Institute) •  49% of outages are caused by human error –  (IEEE)
  • 9.
    80% of ITactivity, time and dollars are spent keeping the lights on - http://www.gartner.com/newsroom/id/497088
  • 10.
    Key wins forautomation: •  Time to delivery/repair –  Zero Touch Provisioning and Replacement of hardware •  Design stability –  Config validation and application •  Avoiding grey failures –  Automated upgrades •  Pro-active fault management –  Remediation and fault isolation Source: facebook / nanog.org Complexity
  • 11.
    Do-It-Yourself Custom development for integration Softwaredevelopment resources & approach DevOps Leveraging existing tools (Puppet, Chef etc) Integration & customization via scripts Turnkey Limited solutions today, focus on legacy models Customers need turnkey solution for cloud-automation! Approaches to Network Automation Cloud Titan High Tech Enterprise / SP Traditional Enterprise / SP 10,000 network elements per engineer 1,000 network elements per engineer 100 network elements per engineer © 2015 Arista - Confidential
  • 12.
    Low Inter-Process Communication AristaEOS Architecture Efficient Publish/Subscribe Publish Notify PIM SNMP BGP Driver SysDB MLAG STP eAPI IGMP Linear Cloud Scale Arista EOS: Resilient Cloud Scale Software © 2015 Arista - Confidential Open &" Programmable Merchant Silicon SysDB CLI SNMP VXLAN" VM" Tracer OSPF STP ASIC" Driver SDN" APIs LAG BGP Modular &" Resilient Universal Cloud" Network Customer Driven" Flexibility Cloud Scale" Economics EOS vEOS EOSPrinciples
  • 13.
    §  EOS NetworkWide Services §  Single network control point for: •  Third party controllers •  Orchestration services. §  Turnkey workflow automation CloudVision Workflow Automation Services Workload Orchestration Services EOS Network Wide Services Open APIs & Partner Access Arista’s Universal Cloud Network Arista EOS CloudVision © 2015 Arista - Confidential
  • 14.
    EOS Devices Workflow Automation Services CloudVisionFramework Management, Monitoring Tools Orchestrators and Overlay Controllers Open APIs Open RESTful APIs EOS Core Services Workload " Orchestration Services CloudVision eXchange (CVX) OVSDB, eAPI, Plugins 3rd Party Integration CloudVision Portal (CVP) © 2015 Arista - Confidential
  • 15.
  • 16.
    Platform for Automationand Visibility across the Network Network-Wide Database Aggregation of Network wide ‘Sysdb’ Abstraction of the " physical network Single integration point to the network More scalable controller integration Realtime data collection and Historic trending State-sync
  • 17.
    Cloud Orchestrators Overlay Controllers www.arista.com Network Services Network" Layer Network Control Point OVSDB JSON Points of" Integration Single point of integration to the physical infrastructure Web-based GUI CLI Platform for Automation and Visibility across the Network Network-Wide Database State-sync © 2015 Arista - Confidential
  • 18.
    Overlay " Controller Scaling ControllerIntegration 18 OVSDB Overlay " Controller Network Layer Controller Layer 10x Improvement OVSDB SysDB State Sync Topology/Device Dependent Topology/Device Abstraction Traditional Approach CloudVision Approach © 2015 Arista - Confidential
  • 19.
  • 20.
    CloudVision Portal KeyFeatures 20 Network Rollback Efficiently revert the network back to a known good state Overlay/Underlay Visibility Physical and logical layout, VNI mapping and performance hotspots Provisioning Turnkey solution for automating network operations Change Management Snapshot of pre- and post- changes for better control Network-wide Upgrade Automate software upgrades across groups of devices Maintenance Mode Production Mode Single Pane of Glass Point and click interface to simplify bulk tasks © 2015 Arista - Confidential Time t Time t-1 ✔ ✗
  • 21.
    CVP - ASingle Interface to Automation © 2015 Arista - Confidential
  • 22.
    CloudVision Core Services Manualbox-by-box operations. Difficult to correlate network-wide. Network-wide synchronization for topology level visibility. No ability to monitor code versions and alert on software bugs Services Dashboards, with proactive notification tools Separate vendor tools for Inventory, Syslog, Management, Compliance & other infrastructure reporting Consolidated reporting view within a single application Typically pair of non- customizable NMS tools built for limited sizing, scale & deployments Big Data Infrastructure fully customizable web GUI Primary Secondary Initial Node1 Object X (Active) Object Y (Active) Object Z (Backup) Node3 Object Y (Backup) Object B (Active) Node2 Object X (Backup) Object Z (Active) Object B (Backup) Node1 Failure Node3Node2 Offline / Failure Node1 Object X (Active) Object Y (Active) Object Z (Backup) Object Y (Backup) Object B (Active) Object X (Backup) Object Z (Active) Object B (Backup) Node1 Node2 Node3 Node1 Node2 Node3 Node4 Node5 Initial Add 2 Nodes Before With CloudVision © 2015 Arista - Confidential
  • 23.
    Cloud-like automation foreveryone Turnkey Provisioning §  Automatically provision switches centrally from a network wide view §  Single tool & view to manage code, configuration, scripts etc. §  Manage switch throughout switch’s DC life cycle using single provisioning process §  Full configuration coverage, including VXLAN §  Configuration compliance validation © 2015 Arista - Confidential Scripting Image Configuration Configurations Images Scripts ZTP/ZTR Existing Approaches With Arista EOS CloudVision
  • 24.
    Change Management §  Performgraceful change controls without impacting applications §  Automatic snapshots pre & post change with delta reporting in network wide view §  Useful for Smart System Upgrades (SSU) or any change control §  Integration into Workflow Tools, such as Service Now © 2015 Arista - Confidential Confidence for Change Controls go- forward Productio n Maintenanc e Roll- back Existing Approaches With Arista EOS CloudVision Configurations Images Scripts ✔ ✔ ✗ ✗
  • 25.
    Availability: Network-wide Rollback § Network-wide scope: Images, configuration, scripts §  Historical state used to recover to previous network state. §  Rolling snapshots during change controls making it easier to roll-back during issues §  Reduce the time-to-recovery during maintenance windows Safety tool for maintenance window recovery A time machine for your network © 2015 Arista - Confidential
  • 26.
    CloudVision: Platform forPartners Partner for NSX Controller for overlay networking VROps for visibility Partner for Converged System Reference architecture, including HP OneView Partner for Controller integration for Transport networks Partner for OpenStack integration Partner for L4-7 services integration Partner for Converged systems Partner for L4-7 services integration Partner for OpenStack integration Partner for Solution integration, including ASM Partner for Windows Server 2016 integration for network controllers Partner for Controller integration for overlay networks © 2015 Arista - Confidential
  • 27.
    Fast path toCloud: Turnkey solution for Automation and Orchestration EOS is now a “Network Wide Service” Opex Reduction: By enabling automation of simple cloud services Extends Arista’s EOS Platforms for Software Defined Cloud Networking 1 2 3 4 CloudVision Summary © 2015 Arista - Confidential
  • 28.
    Thank You -Any Questions? dm@arista.com