Concurrent attach of multiple UEs is unreliable

Problem

attaching multiple UEs through the same gNB at the same time does not work reliably. Basically, some UEs get a PDU session, while others do not get a PDU sessions

Steps on how to reproduce

I used a fairly recent version of the 5gc, tag 2024-june, basic deployment, from the workshop material, see this docker-compose file

To reproduce the problem, use this docker-compose.yaml. It specifies one gNB that connects to the core in the network created by the 5Gc docker compose, and define 10 UEs. The UEs all have different IMSI, all the rest being the same. You need the gnb.conf and nrue.conf. The docker-compose specifies a specific image, but newer images should work as well (I can update if necessary)

What I observe

Out of the 10 UEs, not all get a PDU session. The number seems to be fairly random. A common symptom is that it stops at an NAS uplink message and the core does not seem to follow up.

here is a PCAP that shows, you can filter with ngap.RAN_UE_NGAP_ID == X with X=1,...,10 ul-nas-no-answered.pcapng

Sometimes messages are grouped in a single message. Let me know if that could be part of the problem, maybe we can disable it?

What should happen instead

all UL NAS messages are answered, and PDU sessions are generated