On 30/11/2010, at 6:17 PM, Kevin Blackham wrote: > On Nov 29, 2010, at 15:57, William Warren > <hescomins...@emmanuelcomputerconsulting.com> wrote: > >> I think Karl Denninger has this one called right: >> http://market-ticker.org/post=173522 > > I don't think so. Let's do a little math exercise: > > Comcast charges me $75/mo for my pipe, but let's discount that for bundling, > promos and lower tier services. $30-40 avg ok? > > For that money I get 250GB a month. Let's assume I actually use it - which I > never do, even with Netflix, other VOD, and many habits common to eyeballs - > but for the sake of a number to work with, I do. That's less than 1Mbps > average per month. I'm not factoring in deviation from avg to peak, so I am > going to assume 1Mbps per sub is peak per sub and 250GB is not the average > for the user base.
Average is easy - but the "not factoring in the deviation from avg to peak" is basically ignoring the actual meat of the problem. The human being using a network wants a quite large instantaneous peak during, say, 5pm to 11pm week nights. If you're doing network dimensioning and look at the 5min/avg and assume that's enough then you're wrong and will see packet loss. The more customers the smoother the curve, but at the far edge of the network near the last mile where aggregation starts the difference in cost to cope with this starts to add up when you start doing it cookie cutter style over hundreds/thousands or more sites. Especially if these sites are remote and have power/size restrictions. MMC
