After a period of inactivity (more often that not it is overnight), the networking on UltraSparc machines (Both SMP and Uniprocessor). I am currently running kernel version 2.2.10 - the redhat update to 2.2.5 didn't fix the problem, either. It appears to be picky - we have a SMP machine that the networking doesn't die on. They all use the HME ethernet driver, and have all been configured the same (Except for IP Addresses and simple configuration details)
Can you give me more details about the machines where the problem does occur? What is very useful is: 1) PCI or SBUS 2) Disk controllers used 3) Amount of RAM, number of processors for the SMP cases
The machines are SBUS-based, 256Mb ram each. Single CPU machine is an UltraSparc II. The SMP machines are Dual UltraSparc I's. The disk controllers are standard SUN SCSI ones - they came with the machines. (Sun Enterprise 2's) I can't tell where the problem occurs - all I know is that the ICMP layer is still working (ala ping, etc), but TCP connections never seem to establish their connections. This was tested using telnet. UDP connections are harder to test, and have not been tried. An interesting fact is that the machines that had this problem do not have video cards. This might be a big coincidence, however.
This smells like a TCP bug which has been fixed recently for 2.2.x, would you be willing to try out a test kernel? If so please contact me directly so we can arrange to test things out. BTW, a good way to verify UDP connectivity is to try a hostname DNS lookup via nslookup or host commands.
Ah. An nslookup. I didn't think of that, and can't try it now, either. We also can't try a test kernel, since I was forced by management to install Solaris on them. :/
Since the is no means to make further progress on this bug (the reproducer no longer has the systems available and installed to test a fix or do further debugging) I am marking this as a no-fix until another similar report comes in.