[DPDK/core Bug 1667] Can't configure DPDK poller in busy loop to fix drops

[DPDK/core Bug 1667] Can't configure DPDK poller in busy loop to fix drops

[bugzilla]

[-- Attachment #1: Type: text/plain, Size: 1467 bytes --]

https://bugs.dpdk.org/show_bug.cgi?id=1667

            Bug ID: 1667
           Summary: Can't configure DPDK poller in busy loop to fix drops
           Product: DPDK
           Version: 24.11
          Hardware: All
                OS: All
            Status: UNCONFIRMED
          Severity: normal
          Priority: Normal
         Component: core
          Assignee: dev@dpdk.org
          Reporter: fgiordano@riverbed.com
  Target Milestone: ---

I have an application where I configure multiple NICs with different PMDs and
then I have a simple poller thread that copies packets out of the descriptors
of configured NICs to dedicated rte_rings (one for each port) using
rte_eth_rx_burst. Other threads will then get packets out of the rings and
process packets on a per-port basis.

This poller is configured to run in realtime mode, with high priority on a
dedicated/isolated CPU. It works well and for example on an Intel X710 it can
process 10Gbps/1.8Mpps.

However, sometimes (like 3-4 times a day), the poller can't keep up and drops a
few packets. I have a sleep in my current configuration so I tried to remove it
and have the poller run in busy loop (since it runs in its own cpu and cpu
usage is not a problem). But if I do so, it starts dropping packets like crazy.
That's counter intuitive to me and I would like to know why that happens.

-- 
You are receiving this mail because:
You are the assignee for the bug.

[-- Attachment #2: Type: text/html, Size: 3348 bytes --]

[DPDK/core Bug 1667] Can't configure DPDK poller in busy loop to fix drops

[bugzilla]

[-- Attachment #1: Type: text/plain, Size: 704 bytes --]

https://bugs.dpdk.org/show_bug.cgi?id=1667

fgiordano@riverbed.com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |INVALID

--- Comment #4 from fgiordano@riverbed.com ---
Never mind, the problem seems to be with the scheduling policy (and its
priority) that I use for the poller. With non-realtime scheduling policies, I
can run the poller in busy loop at 100% and not drop a single packet. The bug
is invalid.

-- 
You are receiving this mail because:
You are the assignee for the bug.

[-- Attachment #2: Type: text/html, Size: 2707 bytes --]