Re: Redundant/multiple carriers, rerouting, restoration schemes
"..Gary Winnick said a rule of thumb for international carriers is that at least 30% of capacity should be purchased on a separate network. Is that true? If so, it would seem reasonable to apply the same rule to terrestrial networks, no?"
Teddy, at least he qualified "international carriers" which fits one of the areas where this is especially a good rule of thumb. But what was once simple in this regard, is less simple today, and getting more complex as time goes on. Restoration strategies, that is, not networking in general.
This rule of thumb that he echoed was a more doable measure at one time when all networking was circuit switched (which was also more expensive, I might add, as well). Note, what I am saying is that the one-for-n backing up of routes was more doable in the circuit switched mode, but not necessarily better networking.
-----
[[Begin Sidebar: As a side note, this is a popular tactic of new carriers, pushing this bit of advice in order to aid in kick-starting uptake in their new services. Teleport's (TCG's) main theme during its early years, for example, borrowed from the familiar theme which was shown in a picture that was hung outside of Bob A.'s office in the World Trade Center in NY. It showed a basket containing all the eggs.
On the other side of the door, there was a prop showing King Kong climbing up one of the Twin Towers, with each of the 110 floors representing One Hundred Megabits of capacity which had already been turned up to service, to date. The floor that Kong was on at the time showed their progress as of that day, in other words.
Today, these props seem quaint in retrospect, but at the time they foretold some profound implications for the incumbents, whom, I might add, were still scoffing at the notion that CAPs, as they were called at the time (competitive access providers) could ever go anywhere but down. The year, incidentally, was 1987. End Sidebar.]]
------
Maintaining a diversely routed (or alternate carrier-provided) path at all times is/was a good way to ensure that at least one's critical services were maintained during route and central office outages. But this is an overly simplistic and potentially dangerous prescription today, if followed to the absolute letter of the rule. It is a rule which is best applied with the greatest degree of surety at the end office level, and on very long dedicated routes where a mesh is not readily possible (such as on today's international routes, but even here more diversity is being made available each quarter, portending a time when meshing may become a more viable option for load sharing than a straightforward A-B route pairing).
Again, the rule will continue to hold true between the user location and the first central office. Here, it is not only wise to maintain separate carrier routes which are truly diverse with no overlaps <remind me to tell you about a sidewalk collapse sometime, which took a major bank's retail operations out for a full day>, but to have dual central office coverage, as well.
Beyond the first central office, diffusion into the larger clouds begins to take hold, and it becomes increasinly intractable to maintain such a simplistic set of relationships as A and B routes, at all layers of the stack.
Earlier, it was more manageable. For, this rule held a great deal more weight when all services were circuit switched, i.e., when all circuit routing was done through TDM-based, table-driven switches or multi-route T-1 muxes, with a limited number of predefined options, end to end.
But what have we today? We have Internet Protocol, or IP. IP's routing is dynamic and largely unpredictable, dependent on the vagaries of "other people's traffic flows" and the costs which characterize the remaining routes that survive the disruption... the ones that are still up and working after a break.
What's more, we have fragmentation of roles among ILECs, CLECs, DLECs, ISPs, B-BPs, and all of the same in reverse on the other end of the pipe.
Within the same metropolitan area, for example, one could start with an ILEC, hand off to a CLEC at the transport service level, ship to an IXC which takes you to an ISP or backbone provider across a LATA boundary, and descend at the opposite end in the reverse order, over yet another set of carriers, altogether. Most of these might be providing you with both physical and upper layer services, but some might be providing you with virtual links at the upper layers, only!
It gets extremely blurry at some junctures, and maintaining absolute diversity is near impossible at all points along the way. Fortunately, IP doesn't demand the same degree of diversity one-for-one as predefined routes do, I should add. And this quality of IP is one of its shining characteristics.
A cable in Ohio was cut and the traffic was rerouted through London and Denmark. Don't carriers generally purchase capacity on other networks to insure against this type of event?
Geting back to IP, the determination of these route 'costs' are usually dynamic and algorithm driven, although in some cases they are pre-defined as static routes which are said to be "nailed up" or static routes.
These "costs" are a function of distance, delay, bandwidth, instantaneous congestion, peering agreement terms and other financial factors, prioritization, etc.
Sometimes, often in fact, routing decisions (the actual paths that are taken in meshed IP networks) are anything but intuitive, and could result in traffic going to some strange places, especially on large, heavily congested, networks. Losing a major, heavily trafficked backbone route could, and probably would, do this.
I can't attest to the claims that the referenced ISP's traffic was actually re-routed through London and Denmark - not enough information was provided in the article to validate that - but it's possible that due to such a large disruption on a backbone route, this actually did (or could) happen.
----
Winnick's rule of thumb is a good one. But it doesn't begin to tell the whys and wherefores of the whole picture. Or, his primary motivation for delivering it. Smile, and HTH.
Regards, Frank Coluccio |
