context/reproducible experiment
The issue was encountered when connecting twice a UE to the OAI network without restarting any function neither core network functions, neither the gNB. All the core network and RAN were running on the same machine.
When running the UE for the first time, the user traffic was flowing normally as shows this capture of a icmp ping request and response in the core network:
Then after stopping and restarting the UE making sure that it disconnects from the network properly and connects again with the same ip address, the user data does not flow anymore in the downlink direction:
Capture of ping from the core network to the UE were also performed and confirmed that the problem was about the downlink traffic.
Important points noticed
- We noticed that this issue was only happening if the UE was keeping the same ip adddress from one connection to another.
- Restarting the SPGWU alone fixes the issue for one more connection.
Idea on the issue
From what we observed, we believe that this is probably happening because some structure or object that describe the PDU Session created at first UE connection is partially but not entirely cleaned at UE disconnection so that at second connection the SPGWU does believes it already has some PDU Session ready for this UE and does not setup again properly the downlink channel for the UE. If such structure are indexed by ip addresses, it would explain why changing the ip address solves the issue for one more connection.
Details on our setup
We have bee using the core network, RAN and UE as Snap packages developed in the Mosaic5G project.
The core network and RAN were running on only one machine.
Only one UE was used and connected to the network.
The UE was modified in a way that makes sure it disconnects properly from the network so that it get the same ip address at each connection since this address was properly released.
For more info
This issue was reported by Romain Beurdouche. You can contact me by mail at romain.beurdouche@eurecom.fr for more information on the issue.
context/reproducible experiment
The issue was encountered when connecting twice a UE to the OAI network without restarting any function neither core network functions, neither the gNB. All the core network and RAN were running on the same machine.
When running the UE for the first time, the user traffic was flowing normally as shows this capture of a icmp ping request and response in the core network:
Then after stopping and restarting the UE making sure that it disconnects from the network properly and connects again with the same ip address, the user data does not flow anymore in the downlink direction:
Capture of ping from the core network to the UE were also performed and confirmed that the problem was about the downlink traffic.
Important points noticed
Idea on the issue
From what we observed, we believe that this is probably happening because some structure or object that describe the PDU Session created at first UE connection is partially but not entirely cleaned at UE disconnection so that at second connection the SPGWU does believes it already has some PDU Session ready for this UE and does not setup again properly the downlink channel for the UE. If such structure are indexed by ip addresses, it would explain why changing the ip address solves the issue for one more connection.
Details on our setup
We have bee using the core network, RAN and UE as Snap packages developed in the Mosaic5G project.
The core network and RAN were running on only one machine.
Only one UE was used and connected to the network.
The UE was modified in a way that makes sure it disconnects properly from the network so that it get the same ip address at each connection since this address was properly released.
For more info
This issue was reported by Romain Beurdouche. You can contact me by mail at romain.beurdouche@eurecom.fr for more information on the issue.