Date

Minutes

  • Nick Davey, PLM at Juniper, talking about Tungsten Vision & Futures
  • Hoping to achieve in the future: One of the best, most performant, SDN solutions for telco, service provider, and next gen solutions
    • Improve NFVi deployments
      • Looking to improve dataplane perf
      • Improve vRouter portability
    • Position TF for 5G
    • Optimize TF for Edge Compute
  • Operability will be key
    • "Don't fail quietly"
    • Improve dataplane perf by pushing it further down toward the kernel
  • Deployment is hard
    • For distributed systems in particular
    • Takes a lot of time and expertise to do this right at the moment
    • Will be working on the operator framework within TF
    • More than just an installer; will be a lifecycle management tool for containers, the system, and its state
      • Will rely on other mechanisms w/in k8s to store and communicate that state
    • Hopefully will lead to smoother upgrades and operation
      • Guarantee the results of the operations
  • Telco 5G transformation
    • A lot of edge compute's gonna be needed
    • Vendors are shipping VNFs in containers, so containerisation is key in a 5G environment
    • Looking to get dpdk perf, but for containers
      • Building on multi-interface support in k8s
      • Can connect multiple interfaces to arbitrary networks
    • Make the encrypted dataplane more robust
      • Already have this solution, but it's not particularly visible next
  • Edge compute
    • Support for the OpenStack edge compute project
    • Intro support for real time hypervisors (want to test this? Talk to Nick!)
    • Looking to decouple things to make them more flexible
  • Questions!
    • DH: Anything messaging outside of telco? Cloud native, for instance?
      • Lots of work happening in the cloud native deployments
      • Operator lifecycle management is going to help a lot with this
        • Better fold into the orchestrators we're working with
      • Will be bringing in support for service meshes this year
        • Support SMI, for instance, inside TF
        • Would be able to work for any service mesh
    • RLB: When will we get rid of the dependency on OpenStack Keystone?
      • No hard details on this, but it's definitely a black eye
      • On the radar, but need a plan
    • WS: More about that orchestration extension…?
      • It's about deployment and mostly installers
      • "What orchestrates the TF deployment?"
      • Things like Docker Compose and the like
      • Want an operator agnostic approach
      • Will leave the door open for other orchestrators themselves, of course
    • AA: Is it a problem for TF that k8s doesn't scale well beyond 200 hypervisors?
      • Have a lot of power here in multicluster peering and our seamless overlays
      • Wrap TF around all of the clusters, and it's easier
      • Don't get larger clusters, but get a lot more smaller clusters that act as one
      • More reachability across services 


Action items

  • Nick will send slides ASAP after this so we can attach them to this page