Network end users to pull down 2 gigabytes a day, continuously?

Marshall Eubanks tme at
Mon Jan 8 12:03:46 UTC 2007

Dear Sean;

On Jan 8, 2007, at 2:34 AM, Sean Donelan wrote:

> On Sun, 7 Jan 2007, Joe Abley wrote:
>> Setting aside the issue of what particular ISPs today have to pay,  
>> the real cost of sending data, best-effort over an existing  
>> network which has spare capacity and which is already supported  
>> and managed is surely zero.
> As long as the additional traffic doesn't exceed the existing  
> capacity.
> But what happens when 5% of the paying subscribers use 95% of the  
> existing capacity, and then the other 95% of the subscribers  
> complain about poor performance?  What is the real cost to the ISP  
> needing to upgrade the
> network to handle the additional traffic being generated by 5% of the
> subscribers when there isn't "spare" capacity?
>> If I acquire content while I'm sleeping, during a low dip in my  
>> ISP's usage profile, the chances good that are nobody incurs more  
>> costs that month than if I had decided not to acquire it. (For  
>> example, you might imagine an RSS feed with BitTorrent enclosures,  
>> which requires no human presence to trigger the downloads.)
> The reason why many universities buy rate-shaping devices is dorm  
> users don't restrain their application usage to only off-peak  
> hours, which may or may not be related to sleeping hours.  If peer- 
> to-peer applications restrained their network usage during periods  
> of peak network usage so it didn't result in complaints from other  
> users, it would probably have a better reputation.

Do not count on demand being geographically localized or limited to  
certain times of day. The audience for streaming is world-wide (for  
an example, see

for a few hour slice in the early evening EST on a Sunday - note,  
BTW, that this is for English language content). The roughly equal  
distribution to the US and the EU is entirely normal;  typically the  
peak-to-trough bandwidth usage variation during a day is less than a  
factor of 2, and frequently it disappears all together.


>> If I acquire content the same time as many other people, since  
>> what I'm watching is some coordinated, streaming event, then it  
>> seems far more likely that the popularity of the content will lead  
>> to network congestion, or push up a peak on an interface somewhere  
>> which will lead to a requirement for a circuit upgrade, or affect  
>> a 95%ile transit cost, or something.
> Depends on when and where the replication of the content is taking  
> place.
> Broadcasting is a very efficient way to distribute the same content  
> to large numbers of people, even when some people may watch it  
> later.  You can broadcast either streaming or file downloads.  You  
> can also unicast either streaming or file downloads. Unicast tends  
> to be less efficient to distribute the same content to large  
> numbers of people.  Then there is lots of events in the middle.   
> Some content is only of interest to a some people.
> Streaming vs download and broadcast vs unicast.  There are lots of  
> combinations.  One way is not necessarily the best way for every
> situation.  Sometimes store-and-forward e-mail is useful, other times
> instant messenger communications is useful.  Things may change over  
> time.  For example, USENET has mostly stopped being a widely  
> flooded through every ISP and large institution, and is now  
> accessed on demand by users from a few large aggregators.
> Distribution methods aren't mutually exclusive.
>> If asynchronous delivery of content is as free as I think it is,  
>> and synchronous delivery of content is as expensive as I suspect  
>> it might be, it follows that there ought to be more of the former  
>> than the latter going on.
>> If it turned out that there was several orders of magnitude more  
>> content being shifted around the Internet in a "download when you  
>> are able; watch later" fashion than there is content being  
>> streamed to viewers in real-time I would be thoroughly unsurprised.
> If you limit yourself to the Internet, you exclude a lot of content
> being shifted around and consumed in the world.  The World Cup or  
> Superbowl are still much bigger events than Internet-only events.  
> Broadcast
> television shows with even bottom ratings are still more popular  
> than most Internet content.  The Internet is good for  
> narrowcasting, but its
> still working on mass audience events.
> "Asynchronous receivers" are more expensive and usually more  
> complicated
> than "synchronous receivers."  Not everyone owns a computer or  
> spends a
> several hundred dollars for a DVR.  If you already own a computer,  
> you might consider it "free."  But how many people want to buy a  
> computer for each television set?  In the USA, Congress debated  
> whether it should
> spend $40 per digital receiver so people wouldn't lose their over  
> the air broadcasting.
> Gadgets that interest 5% of the population versus reaching 95% of  
> the population may have different trade-offs.

More information about the NANOG mailing list