Dying Gasp

[mhoppes]

Not a huge feature need but could be useful in diagnosing, but wondering if a Dying Gasp would be difficult to include in the switch.

The idea is basically in the event of a power failure the switch sends an SNMP trap just before completely going dead. This allows for proper disconnection from the network and a stamp as to why the device vanished.

[rockhead]

and the switch 'knows' its losing power how ?

[mhoppes]

Voltage drops. This is actually a pretty standard feature in carrier gear.



Dying Gasp is referenced in section 7.1.2.5.3 of ITU-T Recommendation G.991.2 (12/2003) as the Power Status bit.

Generally, when the main power supply loses power from the primary, it uses the remaining power in the capacitor system to send a final beacon before going dead.

[lligetfa]

the switch sends an SNMP trap...

First we need the switch to be able to send traps. When I asked for SNMP trap support, all I got back was "who needs it?". Nobody before now said they want traps.

viewtopic.php?f=17&t=295&p=1838

[mhoppes]

[sirhc]

The problem here is we are polling the sensors every second which means it could be up to a second before the switch is aware input current is gone and then some amount of time to execute a reaction (this part is small), then to send out a message and considering TCP/IP and the time for far end to receive and respond. You might as well assume you would need 2-3 seconds of power to do something like this and it would require that everything works first try meaning whomever you are sending to. But lets be honest for all this to work the switch would also have to be able to keep the POE ports energized as they would be powering the communications line out.

Now it would come down to how large of CAPs do you have on the logic circuits to be able to hold up without power, and if the POE ports are powering your communications line then your CAPs now have to hold up powering 200-400 watts for 3 seconds.......THESE ARE BECOMING VERY LARGE CAPS!!!!

Neat idea but I do not feel this is possible.

So many good ideas abandoned just because they are not feasible, it is a sad day for ideas today.

[mhoppes]

LOL... OK.. I'll admit I forgot about powering the PoE radios when I suggested this... imagine an airFiber backhaul!!!

[mhoppes]

Wait... wait!

The CPE end-user device is still in design right? Could it be done there? That would allow the NOC to know if the device blew up or was unplugged... could be really helpful in troubleshooting.

[lligetfa]

So many good ideas abandoned just because they are not feasible, it is a sad day for ideas today.

LOL... a dying man will grasp at straws. {groan}

When I was monitoring power/temperature/etc. at the NOC on my last job, I setup very specific conditions so as not to get woken up needlessly. Power blips can look the same as full outages, so I would monitor UPS battery level and the time the UPS spent on battery before raising an alert. I gave myself 20 minutes to get dressed, drive in to town, and to manually start the genset or shed load on the UPS.

It was the same sort of thing with my temperature monitor. I setup a camera with motion detect, tied ribbons to the louvers of the A/C, and had the cam throw an SNMP trap on no motion. I had the trap throw an alarm at the security guard station so they would go reset the A/C. I could then write AND rules that included the room temp so I did not get woken up needlessly but had enough time to drive in before the room cooked off.

Anyway... I digress. Never complain of having too many tools at your disposal. I'd rather have it and not need it than need it and not have it. I used to do central syslog collection and SNMP trap collection so that I could do post mortem RCFA. I also graphed what I could with SNMP and other network tools. It was not all just for post mortem analysis. I baseline monitored and could see emerging trends before it impacted the network enough for others to notice.