Click here to load reader
Upload
tail-f-systems
View
919
Download
3
Embed Size (px)
DESCRIPTION
This is a presentation I gave at Network Automation 2010 in Paris, France.
Citation preview
Version 2023-04-13
Dynamic Service Configuration and Automated Network Configuration
with NETCONF and YANG
Carl Moberg <[email protected]>
@cmoberg on twitter
Introducing Myself
VP Engineering at Tail-f Systems with background in Network Operations (AS1299 and AS3301)
Have had head under the hood of many network products with recurring nightmares from what I’ve seen:• “We didn’t really think about a ‘show config’ command”• “So what you’re saying is that operators normally expects a CLI
on the box?”
On-device OAM is more often than not an afterthought in terms of resources and timing…
…which is the main contribution to our current situation.
The Problem
The problem is: historically no standard (formal or informal) for configuration management
It is a problem because: impedance between service and infrastructure domains can only be solved by large amounts of manual labor, code and risk
How bad is it?
Things that one would expect be possible:• “I’d like to apply this change across two or more boxes and
automatically roll both back if one fails”• “I’d like to back up configuration from one box and store it for
later rollback”
Ways to address this:• Make changes manually using expert resources (opex)• Develop and maintain abstraction in house (direct capex)• Third party to develop and maintain abstraction (indirect capex)
Current Service Automation
Network-oriented services are increasingly:• Complex• High frequency• Costly to fail• Scaling up
Examples network-oriented services:• Enterprise VPNs - network order failures delay customer access• Mobile Backhaul - delay in provisioning impacts cell opex• VLAN provisioning for virtualized workloads - just won’t work
Impact on Service Automation
Service Quality• High failure rates• Network inconsistencies
Provisioning Software• Heavy, expensive device abstractions• Least common denominator-features
Operational Staff• Commonly part of the automation loop• Vendor-specific expertise for common tasks
Introducing a Solution
Operators told the IETF in 2001 that they saw no developments addressing their configuration management protocols
This is documented in RFC 3535 Document work timeline:
• 2002 – Network Management Workshops• 2006 – NETCONF base RFC published• Now – YANG standard document in final phase
NETCONF and YANG Highlights
The NETCONF Protocol:• Distinction between
configuration and state data
• Change validations• Multi-box Transactions• Selective data retrieval• Extensible RPCs
The YANG Language:• Human readable• Formal constraints• Hierarchical• Extensibility through
augmentation
Impact on Service Automation
Before
Service Quality• High failure rates• Network inconsistencies
Provisioning Software• Heavy, expensive device
abstractions• Least common
denominator features
Operational Staff• Commonly part of the
automation loop• Vendor-specific expertise
for common tasks
After
Service Quality• Validated changes• Transactions
Provisioning Software• Device abstractions is
inherent to protocol• Focus on value-added
features
Operational Staff• Tasks that can be, will be,
automated• Vendor-specific expertise
for vendor specific features
Looking Ahead
We are climbing a giant, heading for shoulders:• The network as a schema-driven database?• How dynamic do we want the network to be?• Next useful layer of abstraction?
Thoughts on industry impact:• How open will vendors want to be? Can they?• New breed of third party automation vendors?• Impact on standards?
Conclusions
Current state of industry is legacy, proprietary (or both) and not good enough
Everything possible with programming but cost and risk remains
NETCONF and YANG is a solution with focus on the core of the problem
Interesting times ahead
Thank You!