On 03/02/2015 09:20 AM, Naslund, Steve wrote:
Average != Peak.
What is peak? There is a question for you. If we get all the way down to the
fundamentals of any network, peak is always 100%. There is either a bit on the
wire or not. Your network is either 100% busy or 100% idle at any
instantaneous moment in time. What matters is average transfer rate to the
user experience and even that varies a lot depending on the app in question and
how that app tolerates things like jitter, loss, and latency. It is about
whether data is being buffered waiting for a transmission window and is the
buffer being cleared as fast as it is being filled. A network is engineered to
support some average levels because it would be very cost ineffective to
engineer a wide area network to support peak transmission on all ports at all
times. All studies of network traffic show that it is not necessary to build a
network that way. Our networks are statistical multiplexers in their design
and have been all the way back to the Bell System. You do know that not
everyone can make a phone call at once, right (but who would you call if
everyone was already off hook, get it?)? In fact, it is such a difficult
problem that it is very hard to support inside a single data center class
Ethernet switch. In the wide area, it would be incredibly expensive to design
an entirely non-blocking network at all traffic levels. It could be built if
you want to pay for it however.
::AWOOOOGAAAA:: Strawman Alert!
Nobody's talking about taking poor Erlang behind the barn and shooting him.
We're talking about being able to send upstream at a
reasonable/comparable rate as downstream.
Mike
