@@ -254,6 +254,16 @@ The gNB logs can be followed with:
```bash
docker logs -f oai-gnb-aerial
```
#### Running with multiple L2s
One L1 instance can support multiple L2 instances. See also the [aerial documentation](https://developer.nvidia.com/docs/gputelecom/aerial-sdk/text/cubb_quickstart/running_cubb-end-to-end.html#run-multiple-l2-instances-with-single-l1-instance) for more details.
In OAI the share memory prefix must be configured in the configuration file.
- Sometimes, the nrUE would keep repeating RA procedure because of Msg3 failure at the gNB. If it happens, add the `-A` option at the nrUE and/or gNB side, e.g., `-A 45`. This modifies the timing advance (in samples). Adjust +/-5 if the issue persists.
- This can be necessary since certain USRPs have larger signal delays than others; it is therefore specific to the used USRP model.
- The x310 and B210 are found to work with the default configuration; N310 and x410 can benefit from setting this timing advance.
- For example if the OAI UE uses the X410 and the gNB based on [Nvidia Aerial and Foxconn](./Aerial_FAPI_Split_Tutorial.md) a timing advance of 90 has been found to work well.
## 6.5 Lower latency on user plane
- To lower latency on the user plane, you can force the UE to be scheduled constantly in uplink: `--MACRLCs.[0].ulsch_max_frame_inactivity 0` .