nanog mailing list archives

Re: POE switches and lightning


From: Steven Bellovin <smb () cs columbia edu>
Date: Thu, 13 May 2010 14:52:19 -0400


On May 13, 2010, at 2:24 04PM, Daniel Senie wrote:

While the equipment may well be affected by an EM pulse, if the gear returns to normal after a power cycle, then the 
equipment vendor didn't do their job fully developing the product. A product should be tested to take such pulses and 
should recover provided it has not suffered a catastrophic failure (and in fact it should contain sufficient 
protection to avoid such in most cases).

In working on one particular router in the lab some years ago, I was verifying some software functionality and the 
hardware engineer I was working with reached over my shoulder and used a device that delivered a high voltage spike 
(simulated lightning) to a 10BaseT network port. After I peeled myself off the ceiling (and he stopped laughing), we 
set to work figuring out how to get the device to self-reset after such a strike. One component, an Ethernet hub 
chip, got into a confused state. I was able to detect this in software, so we adjusted the product design so that the 
software could yank the hub chip's reset line.

It's unfortunate that products, both hardware and software, receive minimal quality testing these days. Guess it's 
not a surprise, since buyers seemed to prefer products that were quick to market, with lots of bugs, rather than 
reliability and resilience.

It's not just a matter of "these days" -- lightning is awfully hard to deal with, because of how quirky the real-world 
behavior can be.  I had to deal with this a lot in the 1970s on RS-232 lines -- we could never predict what would get 
fried.  Of course, there was also a ground strikes very near my apartment, where the induced current tripped a circuit 
breaker, blew out a couple of lightbulbs, and and came in through the cable TV line to fry the cable box, fry the 
impedance-matching transformer, and fry the RF input stage on the television...

                --Steve Bellovin, http://www.cs.columbia.edu/~smb







Current thread: