[ILUG] Acturate ping times!

[Ronan Cunniffe]

On Wed, 7 Nov 2001, Niall O Broin wrote:

> Ross wrote
> 
> > I am trying to set up a wireless location system in the range of about
> > 20-30 metres.
> > I plan on locating the person using ping over a wireless
> > connection(802.11b)...
> > The more acturate ping i can get the more acturately i can locate the
> > person.....
> > 
> > Does any one know how to get ping times down to microseconds?
> > Or a program that can do this?
> 
> Sorry to rain on your parade but if I were your advisor on this then right
> off the top of my head I'd have to say "Find another project". Radio waves
> travel ~ 300m / µsec so to get your accuracy down to 10% of your desired
> range you'd need to be able to measure the RTT with nearly nanosecond
> accuracy and there are so many other factors involved in a ping than simply
> radio RTT that I just don't see how it would be feasible. 
> 
> Even with a RTOS, which Linux is not, all you're guaranteed is that you'll
> get a response to an interrupt within a predictable time. The only way to do
> radio ranging is with dedicated hardware which knows what it's trying to do
> and is designed accordingly. The basic principles were worked out 60+ years
> ago - carry on from there :-)
> 

Yeah, this was one reason Nasa cooled on the Beowulf idea.  They were
finding that of the time spent in comms, only 3% of that was on the wire,
97% was spent crawling up and down the TCP/IP stack.  They were looking at
~200mu-sec ping times (2x journey time), which is a complete performance
killer for anything not almost completely parallellizable.  So ray-tracing
is fine, but massively-parallel physics sims over TCP/IP is a sick joke.

This problem is what has given rise to:
a) a rash of 'bare' protocols that have lower overhead than TCP-IP
b) active messaging
c) Intel's Virtual Interface scheme
d) Kick-ass low-latency bus/network hybrids like Myrinet and SCI that do
their protocol stuff in silicon... still 5-10 mu-sec.

You're screwed.  Get yerself a GPS.... :-)

Ronan C.