Commentary
Figure 5-6 typifies the PoP design that ISPs generally use in their PoPs. The smaller ISPs obviously won't have as much sophistication and segregation of equipment, and the larger ISPs will have a significantly larger layout. Many types of access layers are possible—two have been shown by way of example, although other access layers are very similar in implementation.
Point-of-Presence Design
Having established the principle of subdividing the PoP into units depending on function and management, the next important principle of PoP design is to establish a basic layout and replicate this across the entire backbone network. Successful ISPs work on the principle of having only three or four different designs for each PoP. This ensures that the network is not too complex, allowing staff training to be given on each design. It also ensures that the engineering and operations staff are aware that the rest of the ISP's backbone will follow one of the few chosen layouts.
The other advantage of having only a few different designs comes at the point of deploying new infrastructure. Instead of having to take time for the senior design engineers to come up with a new design, the ISP operations team simply can instruct the equipment deployment part of the operation to install a PoP in City A using Design B. The new PoP is delivered with a basic configuration ready for the networking team to integrate into the backbone.
The networking team has no surprises because the design conforms to one of the previously agreed-upon standards. In manpower terms, this means that expensive staff members are not tied up with ordering equipment, building racks, or examining collocation space. ISPs tend to choose one design for an average-size PoP—a scaled-down version of this becomes the small PoP, and a scaled-up version becomes the large PoP. There is little benefit to having any further options; as was shown previously, PoPs are made out of core, distribution, and access layers, so having three designs that incorporate this layering is about as much as can be realistically done without the model breaking down.
A further advantage is that the search for collocation space, or room space to install the facility, is actually simplified. If the design specification dictates a particular dimension with particular electrical requirements, particular air-conditioning requirements, and precise security requirements, it becomes significantly easier to work through a checklist to establish a suitable location. Nothing is worse than going halfway through a build to discover some fundamental design flaw about the facility.
These three issues are probably the key components that most ISPs worry about. There are undoubtedly many others, depending on the depth of detail required, but these are fundamental to designing this particular part of the ISP infrastructure.
Backbone Network Design
Typically two styles of backbone design are in use today. The first employs a startype network, in which the ISP is based in one city and simply installs point-to-point circuits to the other locations. This isn't a wide-area network, as such, but more of a collapsed backbone in the major headquarters hub. The advantage of this style is that management is easier and the network is not complex. Each site has one exit path: to the middle. However, this design is fundamentally flawed when it comes to offering any level of service quality. If any link breaks in the star, the remote site is cut off until the link is restored. If the central node has a problem, the entire network goes down. For this reason, a star network is not used much beyond the initial inception of the ISP business and, in fact, is actively discouraged in some circles these days.
Figure 5-7 shows an example of a network built using a star topology. Notice the single connection from each PoP back to headquarters, where the major facility is. If any circuit fails, the PoP that it connects to the core is disconnected and remains out of service until the link is repaired again. Even more problematic, if the headquarters site has a problem, the entire Internet connection service that this ISP is offering is affected because all external connections from this ISP go through the headquarters. Referring to this figure, it is very easy to see why most ISPs opt for some redundancy in the backbone. This is not a scalable network, nor is it a viable business model.
The alternative is the well-tried and very trusted backbone core design. Major locations are selected as the ISP core backbone, and the rest of the network is distributed around that. For example, an ISP might decide that its six major cities will form the core of its backbone and will build high-speed links between these cities. The rest of the network will be built around this core layer. Some people call this a distribution network because it mimics the distribution design used in PoPs, as mentioned previously. Each site in the distribution layer will have redundant links to the core—one primary path will be a high-band link to the core, and one backup path will go to another distribution layer site. If the interdistribution site link goes down, the primary path functions as normal for both sites. If one primary path goes down, the site has backup through the other site and its primary link to the core.
Figure 5-8 shows an example of how the sites in Figure 5-7 could be connected providing redundancy to each site. Notice that each site has two exit paths, and the network itself has two separate connections to the Internet. Three extra domestic circuits are required, and one more upstream circuit is required. However, the major difference is that the network in Figure 5-8 can be run with 99.5 percent guaranteed availability; it is not at the mercy of any one circuit going down, and it is even resilient against complete site failure and upstream circuit failure.
Figure 5-7. Star Network Topology
Figure 5-7. Star Network Topology
BPOPS
Figure 5-8. Mesh Backbone
Figure 5-8. Mesh Backbone
Looking around the many Internet backbones today, it is easy to see that many ISPs have adopted the latter design. Newcomers to the Internet business are still skeptical about it because they see more than the minimum necessary outlay of equipment and resources. There are trade-offs: If there is a desire to have the network 100 percent available, redundancy is required, no matter how good any vendor might claim that its equipment or infrastructure reliability is. If only 80 percent availability is required, building a cheap network will be just fine. In our experience, most newcomers try to do the former using the latter concepts, and finding an acceptable compromise is often the major challenge.
A further point that often is missing in network designs is that many newcomers are completely sold on the possibilities of the new leading-edge technologies being offered. MPLS is perhaps the most recent example: IP backbone providers (usually ISPs) now can compete with the incumbent telcos to provide VPN technologies to customers who previously were the domain of the telcos only. These new-world ISPs would deploy MPLS on top of their existing network infrastructures—or worse, literally throw together the minimum IP infrastructure necessary to support an MPLS backbone. This strategy always is doomed to failure. The most important point when it comes to designing a network is that it must be robust and foolproof enough that any overlay network such as MPLS can operate reliably and transparently.
ISP Services
The art of putting together the system services required to support the Internet infrastructure probably could fill a book in its own right. Our aim here is to try to explain to the network engineer how to properly position and set up the network portion of the various services that an ISP should be offering to its customer base
Meeting network engineers who dismiss services such as the DNS as "Oh, it's the systems engineers problem" is somewhat disconcerting because the positioning and connection of these services to the Internet backbone is actually quite important.
Network engineers need to be aware of several ISP services. The three most important ones today are the DNS, mail, and news services. Although many principles are the same for each service, it is worth covering each in turn.
The DNS is the public face of the Internet, quite literally. The common complaint from first-time users of the Internet if they can't get to a site is, "The Internet is down." In fact all that could be wrong is that the name isn't in the DNS, or the DNS hasn't been set up properly, or there is some infrastructure problem that renders the DNS unusable. All three scenarios are very common.
The DNS has three important parts: the primary nameserver, the secondary nameserver, and the caching nameserver (or resolver, as some call it). Each attracts a slightly different deployment strategy, and each is really important to deploy properly on an ISP network.
The basic principle for each of these types of servers is redundancy. If one server goes down, the Internet community wants (if not requires) that you have some backup so that name resolution still works. If you have one DNS server and it goes down, the systems using the Internet have no way of mapping any name that you give to an address—nor do they have any way of mapping addresses on Internet infrastructure into names. The former makes the Internet appear down; the latter makes the Internet look like a jumble of numbers rather than the names that we humans can handle more easily.
Continue reading here: Ciscocertified Mail
Was this article helpful?