The embedded network device I am working on will reboot after upgrade. The reboot takes about 10 seconds, but because the interface carrier goes down it engages STP at the switch which can make the reboot feel like its taking as long as a minute which many of our customers find objectionable.

I know that the switch can be configured "portfast" to circumvent this and really is the preferred solution, but we increasingly find that our customers do not have trained IT on staff for this function. Instead they use consultants that charge on a case basis. Or their IT staff is too busy to handle the request. Our device does not perform bridging so STP is not useful anyway.

The hardware is exclusively Intel x86_64 and we use only standard Intel ethernet chipsets (igb and e1000e, 10Gb is coming soon).

So my question is: can carrier be left "up" during a reboot? We use 'kexec' to load the new software directly so we're not going through the BIOS. I don't care about packet loss during this time (it's happening for the duration of STP anyway) and certainly don't want to leave the DMA controller enabled.

Is there a patch available that can do this? We are using CentOS-7.

Hello and welcome to LQ.

Kind of doubt it in your case but you'd have to study this nic onboard more. Even higher end server nic's could drop for a while. Why that switch is so fast may be under rapid stp.

Guess the simple solution might be to place a hub/switch between them??

Thanks, we've done the dumb hub in one or two extreme cases which does solve the problem, but it is not a generic solution for us. We'd have to ship the hub and assume maintenance for it as well, so it's a no-go.

Others here may have some ideas and might chime in.

Does this need to be rebooted? How many times do you have to upgrade this stuff? I was trying to recall the cisco class way back and I'd think there is some way to ignore all that but changing and saving all those switch runs may be a lot of effort.