Skip to end of metadata
Go to start of metadata

You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 11 Next »

There have been security concerns when deploying untrusted workloads using bare-metal containers, which utilize shared kernel from the host and only use cgroups and namespaces for isolation. Kata Containers addresses these concerns by using HW virtualization to isolate each container.  

This additional security layer would allow the support of multiple tenant container workloads in one Kubernetes cluster.   

A telco provider should be able to launch its trusted workloads (e.g. telemetry) using bare-metal containers. In the same Kubernetes cluster, customers would be allowed to bring their application workloads using Kata Containers.  

This will eliminate the need of having multiple VM based Kubernetes clusters for different tenants or the need to assign different compute nodes for each tenant, which should improve the resource utilization of the environment.


Case Attributes 

Description 

Informational 

Type 

New  

  

Blueprint Family - Proposed   Name 

ICN 

  

Use Case 

uCPE Edge Computing (described below)

  

Blueprint proposed Name 

Multi-Tenant Secure Cloud Native Platform 

  

Initial POD Cost (capex) 

Same as ICN, no additional cost. 

  

Scale & Type 

Same as ICN. Minimum of 4 Xeon Servers + 1 Xeon server as bootstrap node.

  

Applications 

Telco trusted workloads and customer untrusted workloads.  

E.g. SDEWAN, EDGX Foundry

  

Power Restrictions 

Same as ICN. 

  

Infrastructure orchestration 

Bare Metal Provisioning

Kubernetes provisioning :  KuD.

Centralized provisioning :  Cluster-API + Provisioning controller (Explore Regional controller)

Containerd for runc and Kata containers.

Virtlet for VMs​.

Service Orchestration : EMCO

MEC framework: OpenNESS

Site orchestrator :  Kubernetes upstream 

Traffic Orchestration within a cluster: ISTIO

Traffic orchestration with external entities : ISTIO-ingress

Knative for function orchestration

  

SDN 

OVN, Multus, Flannel 

  

Workload Type 

Containers, VMs and functions.

Manageability of Bare-metal containers for trusted workloads and Kata Containers (VM based) for untrusted workloads. 

  

Additional Details 

Kata Containers should be deployable across existing Kubernetes clusters using containerd/cri. 

Kubernetes RuntimeClass (from k8s v1.14) and PodOverhead (from k8s v1.16) are features that allow Kata Containers to be selected, managed and monitored with existing Kubernetes tools. 

Kata Containers will not work when used with docker-shim runtime interface. 

 

Contributors

Intel: Adams, Eric (eric.adams@intel.com), Fuentes, Salvador (salvador.fuentes@intel.com), Shinde, Archana (archana.m.shinde@intel.com), Sterrett, Craig (craig.sterrett@intel.com

Verizon: Ravi (ravi.chunduru@verizon.com)

Aarna Networks: Sandeep (ssharma@aarnanetworks.com), Sriram (srupanagunta@aarnanetworks.com



Use case: uCPE Edge Computing

Attributes

Description

Type

New

Industry Sector

Edge, Cloud, Enterprise, Telco

Business driver

There have been security concerns when deploying untrusted workloads using bare-metal containers, which utilize shared kernel from the host and only use cgroups and namespaces for isolation. Kata Containers addresses these concerns by using a lightweight VM to isolate each container.

This additional security layer would allow the support of multiple tenant container workloads in one bare-metal Kubernetes cluster.  

Business use cases

Kata Containers will add hard multi-tenancy capabilities.

By adding Kata Containers and containerd into ICN, a Kubernetes cluster would be able to launch containers using both runtimes: runc for trusted workloads and Kata Containers for untrusted workloads.

A telecommunications provider deploys CNFs such as SD-EWAN and NGFW under a Kubernetes Cluster in an Edge Location. The provider would like to deploy its trusted workloads using runc.

A Customer of the provider deploys normal applications in the same Kubernetes cluster. Even with customer, there could be multiple departments deploying different workloads which would need isolation. For these different workloads, provider could allow to run them using Kata Containers in the same Kubernetes cluster.

Using EMCO, which provides multitenancy capabilities, provider could define edge locations where Kata Containers is available and will use EMCO to redirect and isolate with respect to logical clouds and users.

Business Cost - Initial Build Cost Target Objective

Kata Containers should be able to run in existing infrastructure as long as it has hardware virtualization enabled.

No additional cost from current ICN infrastructure.

Business Cost – Target Operational Objective

Same as ICN.

Security need

Kata Containers runs a lightweight VM, a minimal Kernel and a rootfs for launching containers. This provides more isolation  against attacks from one untrusted container to other trusted/untrusted containers or to the host.


This solution will allow the support of multiple tenant container workloads in a single bare-metal Kubernetes cluster. 

Regulations


Other restrictions


Additional details

Kubernetes RuntimeClass (from k8s v1.14) and PodOverhead (from k8s  ) are features that allow Kata Containers to be selected, managed and monitored with existing Kubernetes tools.


Presentation:



PTL & Committers:

Per Akraino guidelines, the election of the PTL is done via self-nomination from the list of committers.  Anyone that wishes to be PTL just needs to put a 'Y' in the self nomination column.

Elections for the PTL are now open and will remain so until 2 Feb.  If there is more than one nominee for PTL, an election will take place.


Committer

Committer

Company

 Committer Contact Info

Committer Bio

Committer Picture

Self Nominate for PTL (Y/N)

Salvador FuentesIntelsalvador.fuentes@intel.com

Salvador is the engineering manager for the Kata Containers project. Since he joined Intel in 2014, he has contributed to different open source projects for the cloud. 


Y

Eric Adams Inteleric.adams@intel.com


Archana ShindeIntelarchana.m.shinde@intel.com


Ravi ChunduruVerizonravi.chunduru@verizon.com


Amar KapadiaAarna Networksakapadia@aarnanetworks.com


Sandeep SharmaAarna Networksssharma@aarnanetworks.com


Sriram RupanaguntaAarna Networkssrupanagunta@aarnanetworks.com








  • No labels