After some debugging I found that the intermittent data loss was caused by the GSN not properly reacting to Error Indication packets. From GGSN: <0002> ggsn.c:809 PDP(450091417013617:5): Packet received on APN(internet): forwarding to tun apn0 <000d> gtp.c:2681 Packet from 192.168.27.49:2152, length: 24 content: 32 1a 00 10 00 00 00 00 40 8e 00 00 10 00 00 00 05 85 00 04 c0 a8 1b 31 : Received Error Indication <0002> ggsn.c:372 PDP(450091417013617:5): Deleting PDP context <000d> pdp.c:255 Begin pdp_tiddel tid = 5716310714190054 <000d> pdp.c:262 End pdp_tiddel: PDP found When the GGSN receives an error indication, isnt' it supposed to acknowlege the SGSN and attempt to reinitialize the connection? In my case the GGSN simply deleted the PDP context locally and both SGSN and GGSN gets flooded by unknown PDP context error, while the phone keeps trying to transmit data. Tested on lastest git version and lastest release version: osmo-sgsn 1.3.0, osmo-ggsn 1.2.1 Regards, Pierre