Chances are the ISP has some service bandwidth in LA (a Point-of-Presence...or POP). The user has one end (to a CSU/DSU) the POP has the other end. In the POP, there will be some sort of switching device, because it's likely that the data may be dispactched from the LA POP directly to another POP that is closer to the NYC POP.
Whatever the chosen route, the bandwitch from our user's location will be aggregated in some form or fashion. Most common these days would be a huge friggin' router (a T1 aggregator, like the Cisco 10000/(12000?) series, or possibly a Digital Access Cross-connect (a DACS). or maybe even ATM. The bandwidth can be aggregated using any or all of the above, depending on the providers "line of products" (data only, voice & data, whatever).
DACS are widely used, because they can perform a function called "groom & fill." They can take partial "pipes" of bandwidth (like a T1) and "fill" them with other partial pipe's bandwidth, then split it out again somewhere down the line...grooming and filling lets the transport carrier (the "long lines" guys) maximize the utilization of the "Fat Pipes" (like OC48 (2.4G), OC192 (10gig)) that span the country and the world. It might look something like this (sometimes text sucks):
LA - T1,HDLC (User)--> LA (POP) -->router (decides which path to send the data)-->DACS (aggregates the T1 into a DS3,OC3, OC48,OC192,whatever)-->Long Lines (as part of an OC3,48,192)-->NYC (POP)-->DACS (splits the T1 back out) -->router (T1, HDLC)--->end user router (T1, HDLC to Ethernet)-->server.
The "Long Lines" part may be a single hop, or it may be a dozen, depending on the carrier. The actual route may be LA -->New Orleans-Chicago-Toranto-Washington DC-NYC...if you look around the internet, some carriers/providers will publish their route maps. Some carriers will subcontract their long lines to other carriers, some will buy bandwidth from others (like AT&T, SPRINT, Williams, etc).
In either case, the bandwidth may stay as a collective group, or it may be pulled to groom & fill other partial pipes...at each switching node some traffic is split out, some traffic is added...the groom & fill process is dynamic across the span...the goal is to keep the pipes as full as possible to maximize the investment. DACs are also set up for failover switching. If the primary link drops, a secondary link is automatically switched in with little or no interruption. An ATM long lines infrastructure is very similar, has basically the same capability, and is possibly more efficient for pure data transmission. THe structure is basically the same, just substitute "ATM Switch" where it says "DACs."
Since "The Internet" is really just a collection of smaller systems tied together, with a few companies offering "express lanes" from one point to another, routers at each of the switching nodes have a specific routing table that relates to "how to get from here to there." When the traffic reaches an "off ramp," the DACs/switch/router moves it into another pipe heading closer to that destination. Since the long lines paths are somewhat fixed, the routing tables remain fairly static...with adjustments for link failures, load balancing, and traffic management.
Some carriers (like Ameritech in Chicago) have "Super-POPs" or "Network Access Points" (NAPs) that serve a number of other providers. If you dig deep enough on the Ameritech website, they have a list of which local providers are using their NAP out of Chicago.
The end-user's encapsulation pretty much goes away somwhere around the POP. In any given long lines pipe, some channels are going to be what oiginated as Frame Relay, some channels will be traffic that originated as point-to-point, some DSL, some ISDN, some dial-up (PPP). "In the big pipes, all traffic looks the same." Because the destination is known and declared, when the traffic exits the big pipe, it is re-encapsulated according to that path's specification for delivery.
Because of the way the system is set up, you can have 23 locations, all with 56K, send their traffic to a central location who receives it as 23 channels of a single T1. The central location has to have the equipment to process the aggregated, channelized signal...but pretty much any router can do this...or use a MUX to break out the individual channels and route them as separate entities (like some voice, some data).
The real magic here is that the "Fat Pipes" are really strands of fiber, 8.3 microns in diameter (Single Mode Fiber), that's about one-tenth the diameter of a human hair....some "fat" pipe, huh?
The usual disclaimers apply: At best, this is a surface scratching, there are a jillion details omitted. Some details may be bent to reduce an already lengthy post. If anyone else wants to get more specific, have at it.
Hopefully this "fills in some of the gaps."
FWIW, Happy Holidays
Scott
That?s a lot of repeaters. 
Facebook
Twitter