Ultra Broadband: More Than Just Fiber

paolo.franzoi 12/5/2012 | 5:04:05 PM
re: Ultra Broadband: More Than Just Fiber


I don't disagree with your point but....

If the growth is in streamed video, are we not in UDP land.



Duh! 12/5/2012 | 5:04:05 PM
re: Ultra Broadband: More Than Just Fiber

A key point missing in Sterling's analysis was the increasingly important role of CDNs in the ultra broadband system.  For a few reasons: 

First, to the extent that >80% of traffic is still TCP-based, we are reaching the point of diminishing returns on speeding up the access.  TCP goodput (i.e., the user experience) in a nearly lossless network is proportional to 1/sqrt(RTT), which takes into account both link speeds and prop delay.  At 2.4 Gb/s, prop delay from 1 km of fiber is about equal to serialization delay for a 1544 byte packet.   So if fiber length is invariant, it's not hard to see that there isn't much benefit from cutting serialization delay by speeding up fiber links.  Moving content physically closer is a much better investment.

CDN system can significantly reduce the rate of increase in amount of metro and backbone capacity needed, particularly since it tends to damp out traffic transients. This in addition to reducing the amount of traffic through IXPs and peering arrangements.

Not to shill for the CDN guys: they do very well for themselves.  But IMHO,  we're at a point that CDN has to be considered to be integral with the access and metro network architecture.  Performance implications of each are too tightly coupled with the others to allow for siloed design.

Duh! 12/5/2012 | 5:04:03 PM
re: Ultra Broadband: More Than Just Fiber


That's a big 'if'. 

I've seen mixed results.  The Bandwidth Dilemma webinar on Wednesday had a graphic that showed that  the traffic mix is still 70+ percent HTTP.  On the other hand, the Sandvine spring report for North America shows real time (streaming) entertainment at almost 50% of the mix.   If I have time next week, I'll go looking for more data points.  

Regardless of the state of the traffic mix, TCP still remains a significant  portion of the traffic mix, and page latency probably remains a critical factor in customer satisfaction. 


paolo.franzoi 12/5/2012 | 5:03:51 PM
re: Ultra Broadband: More Than Just Fiber

Boy...that's a goodie...let me think about that...nice one!




Duh! 12/5/2012 | 5:03:51 PM
re: Ultra Broadband: More Than Just Fiber

By the way, I had an epiphany over the extended US holiday weekend.

The way the Internet is "supposed" to work, receiving Hosts are expected to use feedback (in the form of detected lost packets) to reduce receive window size, and thus the rate at which transmitting hosts send packets.  In this regime, each flow receives an approximation of a fair share of its bottleneck bandwidth.  However, since participation in this scheme is voluntary, non-conforming flows (including streaming media) starve conforming flows.  Therefore, the pool of digital  bandwidth available to conforming flows is reduced to the bottleneck link capacity, less the sum of rates of the non-conforming flows.  And since end-to-end services offering conditional bandwidth reservation  (c.f., IntServ, RSVP, DiffServ) are not widely deployed,  there is no admission control to allow an operator to size a minimum pool of digital bandwidth for conforming flows.  This means that in a world  dominated by streaming meda, performance of TCP-based flows is going to suffer disproportionately - unless infrastructure build-out more than keeps pace, or schemes like TFRC are widely deployed.

This observation obviously increases the value of CDNs, both by keeping numbers of non-conforming flows out of the core, and by removing potential bottlenecks along the paths of conforming flows.

Sign In