Uploaded image for project: 'OpenShift Virtualization'
  1. OpenShift Virtualization
  2. CNV-12047

Support host-based DPDK with CNV

XMLWordPrintable

    • Icon: Epic Epic
    • Resolution: Unresolved
    • Icon: Undefined Undefined
    • None
    • None
    • CNV Network
    • host-dpdk
    • To Do
    • CNV-12626 - Virtual Network Functions VNF

      Goal

      Link redundancy and aggregation for high-performance DPDK guests

      Background

      Provide support for host-based/OVS-based DPDK for virtual machines running on-top of OpenShift Virtualisation, allowing them to utilise the performance/latency benefits of DPDK without having to rely on the current SRIOV-only solution. The main driver behind this is to allow "telco-like" features to be exposed for virtual machines, and whilst the SRIOV-based solution we have for DPDK today (currently tech preview with 4.8) provides a lot of throughput potential, it doesn't address all of the expectations that customers have, especially if they've come from an OpenStack world. One of the key expectation is for failover/link-redundancy - with OpenStack one can expose a VIF to a virtual machine that has OVS-DPDK acceleration available via the host, and runs on-top of an LACP bond at the host level, meaning the VM running on-top never needs to care about link availability, it only sees a single VIF that has both DPDK support and the benefits of both link redundancy and link aggregation supplied by the host... so both failover and aggregated throughput. With the SRIOV-only approach that we have today, they have to manage the link redundancy themselves from within the guest, e.g. a bond of two VF's, and can only do active/backup bonding (no link aggregation, only the performance of a single VF) due to the upstream switch not being able to provide LACP support for the VF's.

      User Stories

      • As an administrator, I want to offer up DPDK-capable network interfaces for my developers and application owners that provides high throughput and link redundancy out of the box.
      • As a developer/application-owner, I want to deploy my applications into an environment that doesn't rely on me having to manage the link availability from within the guest.
      • As an infrastructure owner looking to move to OpenShift from OpenStack, I need to make sure that we don't see a regression between what we have today with OVS-DPDK and what is being offered with OpenShift.

      Non-Requirements

      • tbd

      Notes

      • tbd

      Owners

      Role Contact
      PM TBD
      Documentation Owner TBD
      Delivery Owner (See assignee)
      Quality Engineer (See QE Assignee)

      Done Checklist

      Who What Reference
      DEV Upstream roadmap issue <link to GitHub Issue>
      DEV Upstream code and tests merged <link to meaningful PR or GitHub Issue>
      DEV Upstream documentation merged <link to meaningful PR or GitHub Issue>
      DEV gap doc updated <name sheet and cell>
      DEV Upgrade consideration <link to upgrade-related test or design doc>
      DEV CEE/PX summary presentation label epic with cee-training and add a <link to your support-facingĀ preso>
      QE Test plans in Polarion <link or reference to Polarion>
      QE Automated tests merged <link or reference to automated tests>
      DOC Downstream documentation merged <link to meaningful PR>

            phoracek@redhat.com Petr Horacek
            rhn-sso-roxenham Rhys Oxenham (Inactive)
            Votes:
            0 Vote for this issue
            Watchers:
            8 Start watching this issue

              Created:
              Updated: