Date
18 Jun 2019
Attendees
VM (Vicky) Brasseur (she/her)
Lisa Caywood
Randy Bias
Will Stevens
Maddison Long
Syed Ahmed
Jill Lovato
Agenda
Start the Recording (started!)
Minutes
RLB with the background on TF
Call to action that the project needs to move forward
Still currently very much in Carrier Land
Mostly carriers or hosting-type companies are leaning in
Need to get out of the network-focused messaging around Contrail & TF
Marketing messaging
Make docs useful for non-network-engineers
Make project easier to deploy with k8s
Why Kubernetes matters
Has effectively won the app platform world for the future
Is standard, with a broad reach
Accordingly, align well with CNCF and that ecosystem
Get mentioned in the same breath as Calico
Get more folks to recognize that TF is a good CNI for k8s
swill: SDN space is in a bit of turmoil right now wrt the market
Some projects are having a challenging time (ODL)
Opportunity to establish leadership in the market
Separate from k8s, but still worth mentioning & keeping in mind
LC: So…appeal of SDN is dead? Or the concept? Or…?
RLB: SDN is VMware messaging to network IT folks
New market is now developers, so need to speak their language
Focus on networking (L4-L7) with Istio (and other service mesh), Envoy, etc.
Aren't thinking of networking/SDN terms
Are thinking of k8s terms
For instance: Tigera & Calico are messaging around security, not SDN
swill: Are in a consumer market now; were in a delivery market for a long time
Based on consumer demand, rather than what the provider wants to build
Developers just want the features. Don't care how it's implemented.
Want to consume functionality.
In the past, esp around SDNs, service providers were in the tech & pushed it up the stack whether consumers wanted it or not
Thinks TF has the capability to support both sides of this
Carriers/telcos aren't going anywhere → Edge computing
TF has bright future in this way
Handles multi-cloud & interconnect between different services quite well
One of only SDNs that handles service chaining in a coherent manner
On other side, is a big gap between service providers and customers
Offering infrastructure, but developers want to consume platform
Service providers have to start delivering PaaS to attract workloads of developers → Kubernetes
Don't need to market as much to service providers since they know about it a bit more
But still opportunity there w/the SDN shakeup
TF well-positioned to be the solution to their problems
Key audiences
Service providers (intentionally vague; many different sorts)
To expose the functionality to be consumed
Value prop to this audience revolves around their consumers
Provide platform that's intuitive and easy to use from a platform perspective
RLB: Carriers don't care quite as much about what the customers want
Use TF as an internal tool to get operational leverage and scale
Are quite comfortable with the direct sales model (Juniper stepping in and providing a networking solution)
Support for open source model, but prefer commercially supported implementations
swill: A lot of orgs are forced to adopt a service provider model even for their own IT
Enterprise application developers
on public clouds
Using k8s as base substrate
Consuming the services
Not traditional on-prem enterprise IT (these are wrapped up in vmware)
Go for the enterprise devs early in their app lifecycle
Start using TF early on, easier to transition to IT
Trojan horse play
As soon as they adopt public clouds for enterprise apps, you step into mult-cloud land where TF plays very well
Public clouds?
No, they're not a target audience for us
But recently they're starting to figure out that multi-cloud is a thing
Have a sympathetic ear because of this, but not a target audience
Hybrid cloud
RLB: TF can drive security and policies even in public clouds
Can be a good story here for basic multi-cloud across public clouds
"Easy failover to GCP if there's an AWS outage…"
Hard to do 2 deployments on 2 different clouds; TF abstracts this away and makes it easier
swill: Example: geopolitically, hybrid cloud is becoming more critical in Canada
Data Sovereignty can become a requirement, simply because data needs to live at home while services are hosted in public cloud
TF has the ability and functionality to simplify this
Define "service provider" → swill
Traditional: telcos
New: a service being operated by a team other than that consuming it
Internal-only
Public
Consumers care about
consuming
tech as a service, not running it
Even if they run it, they'll consume it as a service
Anyone who builds and runs a platform that delivers services
Looking forward for vrouter
There's talk of a next-gen vrouter, separate from TF
RH, Juniper, Intel
Ubiquitous networking mesh
Would ship with the linux kernel, standard interface
Could then select the SDN controller you want
Comparing what TF is today against other things that the audiences are looking at. What does TF offer that the others don't?
VMB: What even are the competitors here?
In k8s ecosystem, it's all Calico (and commercial Tigera)
Cilium
ODL → not really a player as much in our audiences
Canal (Flannel + Calico)
DPDK/SmartNics (but in layman's terms) - marketing the outcome: packet performance acceleration
What does TF do to solve the problems?
Enable use of DPDK
Move network work out of the kernel and into userspace
Allows for much better perf
A feature of DPDK but TF allows you to leverage it without changing your entire architecture
swill believes that with the rise of edge computing then having some sort of network performance optionality will become table stakes
Service Chaining
TF handles better than most other options
Will dig into this a lot more next time
Multi-site connectivity is well handled
Separate from multi-cloud hybrid
Action items
Lisa Caywood
For next time: dig into Service Chaining