16/04/2026 23:11, [email protected]: > From: Pavan Nikhilesh <[email protected]> > > Add configurable mbuf burst size macros: > - RTE_MBUF_BURST_SIZE_THROUGHPUT: optimized for throughput (default 32) > - RTE_MBUF_BURST_SIZE_LATENCY: optimized for low latency (default 4) > - RTE_MBUF_BURST_SIZE_DEFAULT: references the selected profile > > Add meson option 'mbuf_burst_size_default' to select between > 'throughput' (default) and 'latency' profiles. > > Platform-specific configurations can override > RTE_MBUF_BURST_SIZE_THROUGHPUT. > Set to 64 for CN10K which benefits from larger bursts. > > Signed-off-by: Pavan Nikhilesh <[email protected]>
If it would depend only on the HW, I would agree with this config. But it's not, as said already with previous versions by others. It is really an application configuration. I think such config is misleading, and useful only to demonstrate some HW with some specific demo apps. Can we have a constant exported as part of the API to give a tip about the capability of the hardware? Then we could have a smarter decision in the apps.
