[LINK] The PLAN, and broadband speeds?

[Saliya Wimalaratne]

On Fri, Jun 29, 2007 at 06:57:33AM +1000, Richard Chirgwin wrote:
> >>> 
> >>>      
> >>'Scuse me, but 270 milliseconds both ways.
> >>    
> >
> >You're excused, because I'm talking about real-world measured 
> >performance of a few dozen user accounts here, not a calculation 
> >of what it might possibly be under ideal 
> >conditions :) This data was collected and averaged a few years ago
> >and I have no doubt that things are _better_ now - perhaps someone
> >would like to collect current data? 
> >  
> ...Sal, the "real world" test is including all aspects of the end-to-end 
> link (ie, routing etc). It must; because the speed of light certainly 
> didn't change in the last few years... :-)

Hi Richard, 

Not as far as we know :) :)

> The reason I'm indulging in being pedantic is because all IP comms 
> suffer latency and delay in routing; it's inappropriate to compare:
> 
> A) End-to-end session delay in satellite; vs
> B) Physical layer delay only in ADSL.

Ah, but the figures I quoted weren't physical-layer only; they 
were end-to-end for all media. So for the terrestrial links those
figures were of the PPP connection not just the ATM link, for the 
sat it was the first hop before the bird to the end-user node. All
figures include all applicable delays. 

IMHO, directly comparable because it's the 'shortest measurable hop'
for each connection type. Sure, the ends are 'a lot closer' with 
terrestrial tech than with satellite; but that's the whole point, innit? 
If more routers/links/weirdo stream encoders must be traversed to get 
traffic with one tech, you can't leave them out of the benchmark (because
the customer can't :)

FWIW the reason we did all this testing and benchmarking is so we knew 
what would be available to customers in real life - easier to have the
glossy match the world rather than explain why it doesn't. 

I'm sure that higher clockspeeds, faster 'general' technology, different
encoding, and probably most importantly, changes in the ways queueing is 
handled for interactive traffic on sat links mean that slightly-lower 
latencies are probably the norm now. When our tests were done, the idea 
was to get as big a queue as possible per session for maximum throughput - 
now, there are probably multiple separate queues for different traffic types. 

Regards,

Saliya