North American Network Operators Group
Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: BGP Problem on 04/16/2007

  • From: Warren Kumari
  • Date: Thu Apr 19 13:49:19 2007


On Apr 19, 2007, at 12:52 PM, David Temkin wrote:


-----Original Message-----
From: [email protected] [mailto:[email protected]] On
Behalf Of Warren Kumari
Sent: Thursday, April 19, 2007 12:01 PM
To: Robert E. Seastrom
Cc: Leigh Porter; Jay Hennigan; Andre Oppermann; [email protected]
Subject: Re: BGP Problem on 04/16/2007



On Apr 19, 2007, at 10:17 AM, Robert E. Seastrom wrote:



With certain susceptible Sun CPUs which were popular during
the last
sunspot maxima, this was actually demonstrably true (and
acknowledged
by Sun), so don't laugh too hard. 

Yup, Sandia National Labs made a radiation hardened Pentium
and, as far as I remember, was working on a hardened SPARC --
there was also some work done (AFAIR on PPC) whereby 3
processors would run the same instructions and vote on the output...



Thinking of perhaps Resilience? http://www.resilience.com/

God, those things were horrid before they realized that the business
model of assuming "The app will always be OK, the issue will be the
hardware" was completely misguided. I forget what the product was named
at the time, but I'll never forget what a piece of crap it was.

Nah, I wasn't thinking of them -- post-traumatic memory loss allowed me to forget them... There was someone else who's name I have managed to forget who tried to do the same thing through 4 parallel SCSI connectors and fancy OS software -- it was horrendous.. There were 2 motherboards in a case (driven by the same, non-redundant, non- swappable PSU!) and each motherboard had 2 dual channel SCSI cards with cables stretched between the cards. Fancy drivers exposed each board's RAM to the other machine -- there was also a 10Base-2 cable (I'm dating myself here) between the mother-boards for coordination and communication. Every now-and-then your application was supposed to make a system call that would cause the machines grind to a halt and compare their memory -- if there was a difference, the syscall would return non-zero and leave you to figure out what to do about it -- unfortunately because there were only 2 machines voting there was no way to know who was right and who was wrong -- the vendors suggestion was to a: reboot or b: "just choose one and hope you guessed right". Wildly broken system...

I cannot find any of my docs on the system that I was originally talking about, but it was 3 PPC cores in a single package -- there was built in hardware to keep them synchronized and voting. AFAIR, it was a drop-in replacement for the "normal" version of the same device, modulo the power-draw.

Maxwell Technologies makes a triple modular redundant cPCI board with SOI processor and rad tolerant FPGAs that is really nice -- somewhere I think I still have a stash of them...

NB: The above mentions 10BASE-2 and cPCI (which will fit in certain vendors hardware) which *just* managed to keep this on-topic -- hopefully :-)

W


--
If the bad guys have copies of your MD5 passwords, then you have way bigger problems than the bad guys having copies of your MD5 passwords.
-- Richard A Steenbergen