Re: [Paw] Renaming

["Pascal Thubert (pthubert)" <[e-mail]>]

Hello Juliusz:

Actually the theoretical work has been done a while ago, as  a response to your question at the mike when I suggested you dig for MTBF or similar (MTTF, MTBE...). We tend to express a network form of MTBF in a number of 9s. Like five 9s would indicate that the Mean Successful Transmissions Between Errors is 99999. A number of use cases would thrive on 1% loss as long as losses are always well spaced so the missing points can be recovered. This can be true for an industrial control loop, and false for burning a master video. The N 9s are there to make the occurrence of P (say, 4) losses in a row very rare, like a probability of 1/10^(N*P). But the 9s are for unrelated errors and fail to convey a rare burst of errors with a common cause, which is what really hurts. These things happen on radios and that's why we are after diverse paths etc..

There's enough stats for is in https://ftp.automationdirect.com/pub/Product%20Reliability%20and%20MTBF.pdf to take us a long way. Our problem is to migrate the concept from a product to a sequence of nodes and then to more complex structures. This could be refined in an academic paper, I expect that'd be a great topic, and then an RFC would summarize that.

What I'd like to see is an informational RFC that presents wireless reliability in a way that can be consumed by the IETF members, and provides terminology that can be referred in our specs. You seemed curious about the topic; that's quality number one for achieving great results : )

All the best,

Pascal

More links:
http://www.bb-elec.com/Learning-Center/All-White-Papers/Fiber/MTBF,-MTTR,-MTTF,-FIT-Explanation-of-Terms/MTBF-MTTR-MTTF-FIT-10262012-pdf.pdf
https://books.google.nl/books?id=rNLSBQAAQBAJ&pg=PA171&lpg=PA171&dq=reliability+MTBF+MTBE+papers&source=bl&ots=L7BsInfaO3&sig=ACfU3U0kYd1lGRyRR_aD5AET9LOI8ZfZWg&hl=fr&sa=X&ved=2ahUKEwisr-LirLjhAhWPZ1AKHe8yARwQ6AEwA3oECAgQAQ

https://www.dfrsolutions.com/blog/the-challenges-of-wireless-reliability








> -----Original Message-----
> From: Juliusz Chroboczek <[e-mail]>
> Sent: vendredi 5 avril 2019 01:40
> To: Pascal Thubert (pthubert) <[e-mail]>
> Cc: Venkatesan, Ganesh <[e-mail]>; Lou Berger
> <[e-mail]>; [e-mail]
> Subject: Re: [Paw] Renaming
> 
> > We'll also need to provide a definition of reliability and
> > availability and as we discussed, 5nines is probably not the only or
> > even best way to express any of that. Juliusz actually raised the
> > point and I hope he continues on that path and even produces a spec for us.
> 
> I'm flattered you think so, but I'm pretty sure I'm not competent.  You'd need
> someone with more of a background in probability and statistics.
> 
> -- Juliusz