Uploaded image for project: 'RHEL'
  1. RHEL
  2. RHEL-119556

NIC Queue Size Not Persisting After Server Reboot, Causing Traffic Throttling

Linking RHIVOS CVEs to...Migration: Automation ...Sync from "Extern...XMLWordPrintable

    • No
    • Low
    • rhel-net-drivers
    • None
    • False
    • False
    • Hide

      None

      Show
      None
    • None
    • None
    • None
    • None
    • Unspecified
    • Unspecified
    • Unspecified
    • x86_64
    • None

      What were you trying to do that didn't work?

      We are experiencing a recurring issue on Dell XR-8000 systems with Intel NICs. At boot time, the RX/TX queue descriptor size is correctly set to the maximum supported value (4096), and ethtool -g confirms this. However, following a server reboot, traffic through the server via this NIC is significantly throttled. Manually reapplying the queue size using ethtool -G (first reducing to 2047, then restoring to 4096) immediately resolves the issue and restores normal traffic flow. This suggests that the queue size is not being properly initialized or activated post-reboot, despite appearing correctly configured. We have set these via Network Manager under the ETHTOOL_OPTS parameters. As described here:

      https://access.redhat.com/solutions/2127401?band=se&seSessionId=993c2b8e-fc76-40b0-8e47-b342dc2238aa&seSource=Recommendation&seResourceOriginID=f44522ee-795c-49b9-a56e-babaa2e6922b

      We suspect a driver-level or system-level inconsistency in how queue settings are applied or retained across server reboots.

      What is the impact of this issue to you?

      The problem is more of an inconvenience to software developers because they can work around the problem by resetting the RX/TX queue descriptor size with each boot. But this feels like a driver bug and should be reported.

      Please provide the package NVR for which the bug is seen:

      RHEL 8.10 with the -rt kernel.

      How reproducible is this bug?:

      at will. 

      Steps to reproduce

      1. On a Dell XR-8000 system with Intel NICs and the RHEL 8.10 rt kernel, set the RX/TX queue descriptor size to its maximum supported value  of 4096. Confirm with ethtool -g.
      2. Pump enough network traffic through this NIC to saturate it.
      3. Reboot. The NIC saturates at a much lower traffic level, even though ethtool -g still says the RX/TX queue descriptor size is 4096.
      4. Set the RX/TX queue descriptor size to a random lower value.
      5. Set the RX/TX queue descriptor size back to 4096.
      6. Now the NIC supports a high level of traffic again.

      Expected results

      The system should sustain the same level of network traffic after a reboot as before. We should not need to hack the RX/TX queue descriptor size at every boot.

      Actual results

      We need to hack the RX/TX queue descriptor size at every boot to sustain a high volume of network traffic.

              network-drivers-bugs@redhat.com network-drivers-bugs group
              rhn-support-gscott Greg Scott
              network-drivers-bugs group network-drivers-bugs group
              Tianhao Zhao Tianhao Zhao
              Votes:
              0 Vote for this issue
              Watchers:
              4 Start watching this issue

                Created:
                Updated: