No. of Recommendations: 27
Y2Krash and others have made comments about topology of networks and asked useful questions about whether this confers a competitive advantage on one vendor over another. I think the answer is yes but its not as straight forward as it might seem. Before I go any further it is critical to remember that this is a personal view, that is not necessarily shared by my colleagues in Nortel let alone the networking industry as a whole, although many people agree with parts of it.

In an ideal world people would build networks with point to point links between the places they want to connect. Generally this kind of network results in either a mesh because everyone wants to talk to each other or a star because it turns out that lots of things need to connect to the same place. Sometimes you have a ring or a mesh connecting the cores of the stars. Either way you tend to end up with a desired connectivity picture rather like the route map of a large airline.

Unfortuntately in the real world this is not quite practical.
Firstly you generally don't have the physical means to connect everyone directly to everyone so the first step is to replace the mesh with a star where everyone can commnicate. Unfortunately even this has drawbacks as you generally can't lay individual links back in to a central point directly from every source and the central point now has to have near infinite capacity to route or switch comminications.
Finally the central point and the links to it are prone to failure thus you want to have alternate routes and point of aggregation.

Hence the ideal design now becomes corrupted. In the enterprise, that is to say within a building where you have control of cabling, a star of stars is a fairly common basis for design. Generally there is a data center (possibly 2) in which there are servers and other cenrtal equipment. From there lead links to wiring closets and from the wiring closets there are direct links to individual desks. In many cases the link between the wiring closet and the data center(s) is redundant - that is to say it connects using different paths and if you have 2 data centers you may have a link from each wiring closet to both data centers. Typically there are only single links between the user desk and the wiring closet although in areas where failure is costly (e.g trading floors) you may have similar dual homing of devices on a particular desk to 2 separate wiring closets.

Once you leave the building though things become more complex. If you have a campus you may lay fibers between buildings in a meshed form or - depending on the use of the different buildings in the campus - extend the star design by bringing the links from each building back to a central building. However this assumes you have the capability to lay fiber where you want and that your buildings are laid out in a manner that makes this easy.

Generally speaking carriers have POPs or Central Offices and loops of fiber between them. Sometimes the loop goes around in a circle back to the same POP, sometimes it meanders between two POPs. Typically each POP will terminate many of these loops and each loop will normally pass a number of buildings. The POPs themselves will probably be directly connected in another ring, but if the carrier has a major data center in the city it may choose to star links back to the data center. Each fiber loop will probably contain a few dozen strands of fiber (48, 96 or 144) but the carrier deploying the network equipment probably has to pay someone a cost per fiber stand. Thus there is a requirement to serve multiple customers in many buildings on the path with the one fiber.

This means that typically some sort of mechanism to share bandwidth betwene different customers is required. The problem is that the cost constraints mean that the carrier has to serve customers based on physical rings even though the services he wishes to offer are star- or mesh-like.

The big question is how the carrier chooses to treat the loops. The choices are
1) as rings of large bandwidth into which all devices on the ring add or remove traffic as required. In otherwords all nodes on the loop must be able to receive the entire capacity of the ring.
2) as stars where each node on the ring adds/removes a certain fixed bandwidth and is not required to be able to see any other traffic on the loop. However the POPs have to remove and reroute all traffic and may possibly route traffic sent by one node back on the same ring only destined for a different node.
3) as a mesh where nodes on the loops share fixed bandwidth between each other (and possibly also fixed b/w to the head end POPs). The mesh may not be complete (i.e. on the ring buildings A, C and E may communicate with each other as may B and D but A->B traffic goes via the POP).

The same arrangements may also apply to the links between the POPs. Clearly the traffic capacities are likely to be larger in these links.
Which architecture is chosen depends on both the transmission mechanism used AND the perceived majority traffic pattern. In many cases the majority of the traffic is off loop but not neccessarily off metro. In other words it makes sense to have a star logical topology for the access loops to the POPs but a ring or mesh may be more sensible for the POP-POP connections.

In other words there is no WRONG architecture but a set of alternatives all of which can be made to work. In my personal opinion the STAR and STAR of STARs provides the best solution for cases where you do not know the desired data patterns because it is generally possible to start with equal sized pipes and then throw bandwidth at parts of the network where demand is. The big draw back is that the centre of the star (and certainly the center of the star of stars) must be a very high capacity node because it has to route all the data everywhere BUT control is easy and with cunning it is possible to cluster things so that you don't hit the problem today and you just do the classic trickle down effect tomorrow when bigger equipment comes along. There is but one caveat: the STAR must have 2 cores (in other words dual homing). This is something that is "non-trivial" to implement efficiently and without introducing over-complex protocols but which we at Nortel think we've nailed. Won't say any more except that we seem to be signing up carriers based on this idea and hence its a winner.

Print the post  


What was Your Dumbest Investment?
Share it with us -- and learn from others' stories of flubs.
When Life Gives You Lemons
We all have had hardships and made poor decisions. The important thing is how we respond and grow. Read the story of a Fool who started from nothing, and looks to gain everything.
Community Home
Speak Your Mind, Start Your Blog, Rate Your Stocks

Community Team Fools - who are those TMF's?
Contact Us
Contact Customer Service and other Fool departments here.
Work for Fools?
Winner of the Washingtonian great places to work, and Glassdoor #1 Company to Work For 2015! Have access to all of TMF's online and email products for FREE, and be paid for your contributions to TMF! Click the link and start your Fool career.