RocketMan, Curtis, All,
re: network bean counting, RM, your post got me to thinking again...
It's the Devil's time, again.
===========
Many, if not most, of the dedicated private line networks -- the ones which were based traditionally on nailed up T1 and circuit-switched links, or using Frame Relay or X.25 services, are now moving over to IP clouds. The former private lines were not very efficient, it turns out, for reasons which are now legion.
The newer nets may be considered intranets, extranets, VPNs of late, whatever. Let's lump 'em all together for discussion's sake. No, I am not conceding that all private IP nets are part of the open Public Internet ;-) but for the sake of discussion...
The overwhelming majority of these newer network forms share a certain common characteristic: They are, in the main, or for the most part all going in as substitutions or replacements for the earlier forms of networks which I enumerated above: The T1-based private line networks.
These are, to a large extent at this time, mainly "substitutions," in other words, or replacements, I submit, for the previous networks that were serving the same purposes. And they use less bandwidth, not more bandwidth per function served, for reasons stated above. Namely, because IP networks are more efficient.
One of the things that is actually changing is the means of allocation of high capacity bandwidth swaths in order to aggregate the IP flows. Whereas, in the past, equivalent or even greater numbers of flows using far more bandwidth per flow, was consumed in satisfying the same end games. In the older model, smaller denominations of line widths were used, in the form of discrete tributaries at subrates, DS0s, fractional T's and T1 rates. And less often, T3s.
This was the way it was prior to lumping thousands of IP flows together in a fewer number of OC-3s, OC-12s, OC-48s, etc.
In the past, however, SONETized networks were grown to accommodate the lower speed tributaries, much more readily than the larger pipes. Witness the abundance of Digital Cross Connects in existence which are designed for distribution to users at the T1 rates and below. Grooming and parsing of these bandwidth denominations on a switched basis was the rule of the day, whereas today everything can be lumped and parsed at the upper layers through routers.
What's my point here? My point is that an enormous amount of a new bandwidth "flavor" must go in at this time in the way of OC-48 tributaries to satisfy the Internet Protocol, because IP does not take kindly to getting shredded through DACS machines.
For that reason, the older infrastructure that is still in place and usable, to a large degree, and still very substantial, is shunned by the ISPs, and for good reason. It just doesn't work for them.
Therefore, the need for newer and larger pipes (although much fewer, still, than the number of the older virtual narrow pipes) that meet the ideal profile dimensions which are more suited for the larger IP flows. The older carriers have more of the older stuff [T-1s and below pipe fittings] in place, obviously, and the newer carriers are leapfrogging them with, you guessed it, the newer stuff [water mains the size to satisfy the State of Texas during a drout].
And large flows are facilitated only by fiber, and to maximize the numbers of flows you get out of a fiber, the only way to do it today is through DWDMs. And that is what the newer guys are putting in at a feverish pace, in order to size their nets to allow for the substitution and displacement effects that I've cited above.
Getting back to the substitution effect, as such, the newer enterprise IP nets will logically top out at some point, when the migrations of applications to the newer IP format are finalized.
This end-state condition (when the migrations are over) will show up as a plateau in growth for this sector of activity at some point, and the plateau level will be sustained or show a more moderate rate of growth than previous predictions have suggested, until other forces that will drive bandwidth demand become prevalent. These will most likely be when m-m begins to pick up speed, and as other new reasons become apparent for using the 'net.
Notwithstanding, there will be increases in the private-citizen internet access component, and it will continue to grow as the numbers of users rises, which only makes sense. But these growth rates will pale in comparison to the bandwidth impact being offered by the enterprise contributions at this time.
In fact for every facet of internet activity, what one could actually do is predict a similar topping out period. This goes for CM rollouts, DSL deployments, etc. The reverse of extrapolating growth numbers, in effect.
How long will the individual plateaus last? What are the factors that will dictate these hiatuses? I don't know what they are in real terms, but I have some ideas.
My point is that even though IP makes it more affordable to start new networks now, the preponderance of all new ones actually going in at this time are involved in a displacement effect, rather than representing incremental growth where overall bandwidth is concerned. Possibly just to the contrary! [uh, ooh... here's where it gets dangerous...]
Indeed, if IP is as efficient to the overwhelming extent that we have given it credit for, at least far more efficient than traditional forms of switched circuit networking, then we should actually be seeing an overall decrease in the requirements for overall bandwidth which would be needed to satisfy the world's total networking needs. No?
Stated another way: Even though there are other forces at work that will push bandwidth demands through the roof at some point in the future, many of the specialized enterprise and government networks that are NOW rising quickly in overall resource requirements, including bandwidth usage, will actually top out at some point when the internal needs of those individual private entities are satisfied, and these should NOT be used for extrapolatory purposes.
The Devil Has Spoken.
Best Regards, Frank Coluccio <<shields up>> |
