>> What is the difference between the NJE keepalive and the RSCS >keepaliv=yes? We >> have tried both =yes and =no to no effect. We have been running with >keepalive >> turned on for 48 hours. The same errors show up like clockwork when the >link is >> idle for an extended period. > >I'm not aware of any NJE-level keepalive. > >The time for speculation is past. Gather your evidence and let it speak >to you. BE the CSI. > >What do the network (not host-created) packet traces show at both ends? >Are the keepalives going through? If you see TCP RSTs coming in that the >other end didn't send, then the game's afoot and the challenge is, then, >to locate the offender. > >For fun, I did a search and saw z/OS APAR PK16058: KEEPALIVE PACKETS NOT >ARRIVING AT PARTNER SYSTEM, which tells me it is possible that keepalives >are reaching z/OS, but not getting back to z/VM, and z/VM's own keepalive >support is terminating the connection. >
There is no TCPNJE keepalive architected interface. Other NJE architectures have defined a way to determine if the link is still alive, but this was not included in the TCPNJE definition. In addition, the RSCS keepalive= parm is not RSCS doing keepalive activities, rather it turns on a bit in the socket to request TCP/IP implement keepalive for the connection. Best Regards, Les Geer IBM z/VM and Linux Development