Bringing DevOps to Routing with evolved XR: an overview

16
Bringing DevOps to Routing with Evolved XR Akshat Sharma, Technical Marketing Engineer Web Solutions

Transcript of Bringing DevOps to Routing with evolved XR: an overview

Page 1: Bringing DevOps to Routing with evolved XR: an overview

Bringing DevOps to Routing with Evolved XR

Akshat Sharma, Technical Marketing Engineer

Web Solutions

Page 2: Bringing DevOps to Routing with evolved XR: an overview

DevNet-1089

DevOps

Breaking the “Ops” Dichotomy

2

NetOps

Gateways,

LERs, LSRs

Servers, TOR switches, Spine/leaf switches

Page 3: Bringing DevOps to Routing with evolved XR: an overview

DevNet-1089

30% Growth is Forcing Web Providers in Two Directions

“Scale breaks hardware. Speed breaks software. Speed at scale breaks everything”

Adrian Cockroft, Technical Fellow, Battery Ventures (formerly Cloud Architect at Netflix)

Speed

Scale

Page 4: Bringing DevOps to Routing with evolved XR: an overview

DevNet-1089

What does speed at scale look like?

15 Minute Truck to

Live Traffic

Deployment time for

DC Pods

Ticket Systems

Planned Outages

Time to Deployment measure in months

NetOps

DevOps

Page 5: Bringing DevOps to Routing with evolved XR: an overview

DevNet-1089

Some more examples of Speed @ Scale

• Deploy services 300 times an hour

to an average 10,000 servers

• 15K engineers working 4K projects with 5500 commits and 75M test cases run every day

• Build, test and deploy new code in ~60 min

• Deploy bug fixes and internal changes twice a day, and new features once a week Source:

NANOG ‘15

AWS re:invent ’15

GCPNext ’15

Page 6: Bringing DevOps to Routing with evolved XR: an overview

DevNet-1089

DevOps

Taking Stock of the tools

6

DPDK Open NSL

Silicon

Abstraction

Infra

Software

Service

Orchestration,

Automation &

Analytics

Tupper

ware

GEARD

Data

Stores

App

Servers

Services Front End

P4

NMS

NetOps

Page 7: Bringing DevOps to Routing with evolved XR: an overview

“Web Scale” 2006 - 2015(public statements by web companies)

2014: Server to Admin 35,000 : 1

2012: Server to Admin 10,000 : 1

2015: Server to Admin 50,000 : 1

2013: Server to Admin 20,000 : 1

2008: Server to Admin 3,000 : 1

Traditional Data Center Avg: 50:1

Standardization of Environment (standard scales)

+ Automation & Auto-remediation (via Web APIs)

+ Predictive Analytics (using Big Data)

Massive Scale

NetOps is now at

3,000:1 in Web

Page 8: Bringing DevOps to Routing with evolved XR: an overview

Traditional Monitoring is Showing its Age

SNMP

Syslog

CLI

Data created here… …reactive analysis here

• Incomplete and inconsistent view of infrastructure

• Multiplicity of formats and mechanisms needed to be normalized and correlated

• Underlying approach not scaling to meet contemporary needs

Page 9: Bringing DevOps to Routing with evolved XR: an overview

DevNet-1089

SW Toolchains to Reduce OpEX and Increase Service Velocity

Day 1

CONFIGURE

Day 2

MANAGE & OPTIMIZE

Day 0

INSTALL

* Source: Google

Software

Tooling

Visibility

Automation

Control

Web Scale

Operations

• 30k Configuration

changes per month

• 20k Command Line

issued and scrapped

every five minutes

• 8 Million objects polled

every five minutes

• Operational efficiency

• Optimized utilization

• Service agility

Benefits

Page 10: Bringing DevOps to Routing with evolved XR: an overview

DevNet-1089

IOS-XR 6.0 Operational Enhancements

Vis

ibili

ty &

Tele

metr

y

Evolv

ed p

rogra

mm

abili

ty

Applic

ation H

osting

Fle

xib

le P

latform

& P

ackagin

g

IOS-XR 6.0

10

Page 11: Bringing DevOps to Routing with evolved XR: an overview

DevNet-1089

Example 1: Router “State” enforcement

11

The Traditional Approach :

Enforce state through periodic “Commit Replace”

10000 line config file

Page 12: Bringing DevOps to Routing with evolved XR: an overview

DevNet-1089 12

Example 1: Router “State” enforcement

Logstash KibanaElasticsearch

The DevOps approach:• Config Management tool to enforce

intended state

• Stream real-time Telemetry data to detect “state” discrepancy

• Remediate “diff” through a chef/puppet/ansible run and/or raise alerts!

Enforce Partial/Complete State

Page 13: Bringing DevOps to Routing with evolved XR: an overview

DevNet-1089

Example 2: Testing and Deployment

13

Weeks of Testing on Physical hardware: Expensive, non-representative

Multiple Step “Manual” Deployment Process:

• Load Images, config manually

• Verify connectivity

• Remediate errors manually

The Traditional Approach :

Page 14: Bringing DevOps to Routing with evolved XR: an overview

DevNet-1089

Example 2: Testing and Deployment

14

Automated Testing of PI features on

Virtual Hardware:

• Representative Topologies

• Cheaper

Automated Deployment process

• Auto-provisioning fetches images/config automatically

• Configuration Management tools enforce intended state

The DevOps Approach :

VV

V

V

V

V

V

Autoprovisioning

Page 15: Bringing DevOps to Routing with evolved XR: an overview

DevNet-1089

Dive Deeper?

15

• BRKSPG-2069: Introduction to IOS-XR 6.0

• LTRSPG-2517: IOS-XR 6.0 hands-on lab

• Devnet-1086: Extending Streaming Telemetry

• Devnet-1088: Developing on IOS XR

Page 16: Bringing DevOps to Routing with evolved XR: an overview

Thank you

16