?c2go

[Hal Murray]

esr at thyrsus.com said:
> I still don't understand that suggestion at all.  Would it be implementing
> the Byzantine time=sync algorithms? If so, howe is it simple?

No.   That's what makes it simple.

> I guess I need to see something more like a functional spec for your
> concept. It sounds like you might have something interesting in mind, but I
> am completely failing to follow your thinking. 

Assume your local clock is good.  Use it.

All it does it answer NTP requests.  Just receive a packet, drop packets with 
invalid fields, fill in the appropriate slots and send it back.  If there are 
fields that can't be filled in correctly, fake something.

That should be close to a server running off a local refclock.  Maybe even 
identical.

It gets a chance to poke in lots of interesting corners without getting 
overly complex.  You have to do network I/O and byte swapping.  You have to 
interface to c code.  You have to convert from Unix time to NTP time.  ...

If that's done as a hack, we can measure performance.  If it's none well, 
with a multithreaded version, it might be appropriate for busy servers.  I'm 
thinking of the NIST level of traffic.

It might make an interesting platform to experiment with crypto.


-- 
These are my opinions.  I hate spam.