The AutoCon 0 day two (Tuesday, 14 Nov) keynote speaker was an engineer who's been talking about "the self driving network" for at least a decade already. Kireeti Kompella is the CTO, PSD at Juniper Networks. Before taking on his current role, he served as CTO, SDN at Juniper Networks. Previously, Kompella was CTO at Contrail Systems, which was acquired by Juniper in December 2012. Prior to joining Contrail, he was CTO and Chief Architect, Junos at Juniper Networks.
Kompella has deep experience in Packet Transport, large-scale MPLS, VPNs, VPLS, and Layer 1 to Layer 3 networking, and has been active in the IETF, as former chair of the CCAMP Working Group and as author of several Internet Drafts and RFCs (in the CCAMP, IS-IS, L2VPN, MPLS, NVO3, OSPF, and TE WGs). Prior to joining Juniper in 1997, he worked on file systems at NetApp, SGI, and ACSC (acquired by Veritas). At heart, Kompella is still an engineer and a coder, and loves talking to ASIC folks.
Kompella received a bachelor of science degree in electrical engineering and a master’s degree in computer science from IIT, Kanpur, and a PhD in computer science from University of Southern California, specializing in number theory and cryptography. He holds 46 issued patents.
2. Forward-Looking Statements
This presentation contains forward-looking statements within the meaning of Section 27A of the Securities Act of 1933, as amended, and
Section 21E of the Securities Exchange Act of 1934, as amended, which statements involve substantial risks and uncertainties. Except for
historical information contained herein, all statements could be deemed forward-looking statements, including, without limitation, Juniper
Networks’ views concerning our business, economic and market outlook; our expectations with respect to market trends; our product
development; the strength of certain use-cases and customer segments; the introduction of future products; the strength of our solution
portfolio; the timing of recovery from COVID-19 on customer demand and resolution of supply challenges; and overall future prospects.
Actual results or events could differ materially from those anticipated in those forward-looking statements as a result of several factors,
including: general economic and political conditions globally or regionally; the duration of the effects of the COVID-19 pandemic; business and
economic conditions in the networking industry; changes in the financial stability of and overall technology spending by our customers; the
network capacity requirements of our customers and, in particular, cloud and communication service providers; the timing of orders and their
fulfillment; manufacturing and supply chain constraints, changes or disruptions in our business operations caused by, among other things,
armed conflicts, cyberwarfare, political tensions, natural disasters and climate change; availability of product components; delays in scheduled
product availability; adoption of regulations or standards affecting Juniper Networks’ products, services or the networking industry; the impact
of inflationary pressures; executive orders, tariffs, governmental sanctions, changes in laws or regulations and accounting rules, or
interpretations thereof; and other factors listed in Juniper Networks’ most recent reports on Form 10-Q and 10-K filed with the Securities and
Exchange Commission. These forward-looking statements are not guarantees of future performance and speak only as of the date of this
presentation. Juniper Networks undertakes no obligation to update the information in this presentation in the event facts or circumstances
subsequently change.
3. A. 1 = 1
B. 1+1 = 3
C. 1+1+1 = 7
D. Holistic Automation
E. Self-Driving Networks:
Still the Holy Grail?
F. What’s Your Vision?
5. 1 = 1
Choose a pressing problem
Identify data needed to solve it
Process the data
Take relevant action
^
operational
6. Examples from Driving
Data: determine lane markers
Process: is car centered?
Action: adjust car position
Data: determine speed
Process: compare with “intent”
Action: speed up/slow down
Cruise control “Lane keep”
Action is taken automatically
However, human must monitor car
7. Deploy Basic Automation à Human Acts
Data: device syslogs
Process: dedup, filter
Action: root cause analysis
Data: e2e bandwidth stats,
Process: time series analysis
Action: capacity planning
Data: device telemetry
Process: identify hotspots
Action: rectify anomalies
Data: ingress stats
Process: anomaly detection
Action: determine whether
flash flood or DDoS attack
Data: sample ingress traffic
Process: send to IDP engine
Action: drop if malicious
Data: optical link errors
Process: identify fiber kinks
and laser malfunctions
Action: move traffic
Data: topo, e2e b/w, link b/w
Process: identify congestion
Action: rebalance traffic
Data: topology, e2e b/w
Process: traffic engineering
Action: move traffic
9. 1 + 1 = 3
Choose two related problems
Identify data needed for both
Process the data
Take unified action
10. Example from Driving
Data: determine speed, find lane markers
Process: compare speed with intent, car position
Action: manage speed (brake/accelerator), center car (steering)
Enhanced cruise control
Action is taken automatically
Again, human must monitor car
11. PCE controller knows topology,
current network state,
e2e flow bandwidths
à computes paths for e2e flows
Traffic Engineering
If topology changes:
àcompute new paths
for affected flows
à “move traffic”
Original slide from Julian Lucek
(2019), modified slightly here
12. Congestion/Gray Failure Avoidance
PCE Controller is told via
Streaming Telemetry how
much traffic is on each link
So, it automatically moves
away some LSPs from the
congested link
Similar action can be taken
if a link has “gray” failures:
not quite down. Controller
must be told to avoid link
Original slide from Julian Lucek
(2019), modified slightly here
15. 1 + 1 + 1 = 7
Set of related problems = use case
Identify data needed for all problems
Process the data
Take unified action (typically via a workflow)
16. Example from Driving
“Hands-free driving”
Action is taken automatically
Car monitors human (!)
Data: determine speed, distance to next car & lane markers
Process: compare speed/distance with intent, car position
Action: manage speed (brake/accelerator), center car
(steering)
Convenience!
17. Use Case: Assured Onboarding
1. Is the device genuine?
2. Is the device correctly connected?
3. Update the software to the desired release
4. Configure the device
5. Is the device configured correctly?
6. Does the device have appropriate reachability?
7. Monitor the device on an ongoing basis
8. How is the device doing compared to its peers?
9. Is the device still healthy?
10.Bring on more devices (or change existing)
11.Still connected appropriately?
18. Use Case: Transport Network Slicing
eMBB
mMTC
URLLC
The desired SLOs must be met,
the experience must satisfy
customer (or app)
Intent
Experience
Network
View
Device Configs
Set up Probes
and Metrics
Intent Compiler
(connects Intent to Experience)
topo filters, slice
aggregates, paths,
CT mapping, FAs,
PHBs, bw engg
25. Network Digital Twin
“Planner” on steroids
Platform for training
Exploring via Mixed Reality?
“What happened” analysis
“What if” scenarios
Devices (and software)
Topology (planning)
Data Plane (traffic)
Control Plane (protocols)
Management Plane
26. CI/CD Pipeline
Software-style discipline for automation changes
Automation as code: snapshot and version
Commit à run the tests (on digital twin?)
Success: deploy or Fail: roll back
28. Self-Driving Networks:
Still the Holy Grail?
8 years since the original vision
Seemed like science fiction then
We’ve learned a lot
(streaming telemetry; machine learning for networks, …)
We’ve come a long way: on the cusp now
Time to step back and ask: what’s next?