mirror of
https://gitlab.eurecom.fr/oai/openairinterface5g.git
synced 2026-07-13 04:30:28 +00:00
8556a1488c
Reduce number of mutex locking in MAC->RLC direction Currently, the scheduler locks RLC for each operation. Concretely - for RLC status indication (to get amount of bytes per LC), it locks per LC - for RLC indication (UL traffic), it locks for each MAC SDU of which there can be multiple in a TB - for RLC data req (getting data from RLC to put into a transport block), it locks for each LCID and PDU to get. This can have a certain overhead, because we need to re-lock RLC multiple times instead of once (per operation above). In this MR, rework the interfaces such that it is possible to get all information/data in a single lock-unlock cycle, which should also help with "spikes" of scheduler times (in some scenarios, it seems the scheduler takes longer than 0.5ms, and while it's not only locking, re-locking might contribute), and an increase of number of UEs (e.g., instead of doing 3 * N UEs lock cycles for getting status of SRB1, SRB2, DRB1, we do only N times). Concretely, for each UEs RLC status indication, we lock once, only once per LCID when filling a TB (in particular for LCID 4, on high throughput, it might be many times), and only once per UL TB (for the common case of UL data). Minor fixes include the increase of the Aerial VNF thread priority, and reducing the maximum RLC indication size to reduce excessive computation (we can't have 1MB in one TB, it is too much).
This directory contains the NFAPI code.
It comes in two parts:
1 - open-nFAPI
This is a clone of the github repository
(https://github.com/cisco/open-nFAPI,
commit b3bc579b1697eab829d5d8a2de59c93a61b88fa4).
The patch open-nfapi.oai.patch has then been applied.
2 - oai_integration
This is code written by David Price from Cisco to integrate
open-nFAPI into OpenAirInterface.