I work a lot with ETSI NFV, NFV-O and Automation for Applications running in VMs… CloudNative requirements for Telco Cloud triggers the need to have Containers… Some vendors are considering Containers over VMs to support Containers but without changing the Orchestration for VMs (NFV-O) and all integration with BEs… My question is a bit naïve… In case we have apps running on containers and apps running on VMs, how we co-exist the different Orchestrators without overloading the IT back-end? LCM, Scaling-in/out is easy to understand the need to have two orchestrators managing their own environment… But, from Telcos IT perspective, if there´s need to have VMs and Containers and NFV-O and Kubernetes, how to minimize the impact and integration requirements with IT back-ends with multiple Orchestrators handling automation?
Found some info from ONAP´s blog… But doesn´t help with the integration and overload for IT systems: https://www.onap.org/blog/2018/05/04/how-service-providers-can-use-kubernetes-to-scale-nfv-transformation
Hi! We took a stab at this question during office hours, have a look here:
Hi guys! Tks for your time trying to help me out… Telco VNFs provisioning is not a straight forward process… And the environment is Stateless… Telco VNFs Provisioning is not integrated with NFV-O! We make the services and users provisioning first then use NFV-O to scale-in/out, to handle the life cycling… If the Operator wants an Automation, then, there´s need to integrate the NFV-O with their IT supporting Systems… I´m seeing some solutions, for a Stateless environment that considers a Middleware deployment in between the Multiple Orchestrators (NFV-O, Kubernetes and SDN-O for layer2) in order to simplify the integration with their IT systems.
Do you have more info about VMs management in a Kubernetes way?
@icoeli The best resources that I know of are the kubevirt and virtlet projects directly. This scenario is still somewhat of an edge-case…there really isn’t a lot of documentation or examples in this area unfortunately.
If you do wind up finding a solution that works for you though, I do think the community as a whole would benefit from a post or summary your findings Would definitely make for a good talk to two to boot!
Thank you, Bob… I´ll definitely post here my findings… Thank you, again
Something else that just came to mind. It doesn’t necessarily help with the orchestration side, but we’re definitely seeing a lot more work being done on integrating Kubernetes with SRIOV and VNF/NFV functions.
If you haven’t already given it a look, take a look at the multus CNI driver. It lets you use and configure multiple network interfaces for each pod and they have some good links and use-cases in the repo. I haven’t had too much time to play with it, but we’ve been looking at it as a possible option to carve up and attach infiniband devices to pods for HPC workloads.