libvma
libvma copied to clipboard
issue: 2336523 Return RX pbufs to cq_mgr instead of global pool
Global buffer_pool lock leads to a lock contention in multi-threaded applications. We can mitigate it by using ring per core and returning pbufs to respective cq_mgr cache.
Rebased on top of vNext and updated with the latest version.
@pasis please update with addirional fix https://github.com/Mellanox-lab/libvma-pro/pull/22/commits/4ae9b6e7a49bda85d18f00cb5d10a66ff861342f
Can one of the admins verify this patch?