Dial backup is an important feature in a reliable WAN design. If the primary link to a remote site fails, dial backup links can ensure that you don't lose all connectivity. Of course, the dial backup link will usually have significantly lower bandwidth than the primary link. However, the principle advantage of using a dialup connection for backup is that the link will only connect when required. The rest of the time the connection is down, which usually saves money, because you only pay for the access and avoid the connection charges.
The examples in this chapter are also useful for WAN designs in which the dial links are used as the primary connections. There are two common examples of networks like this. The first are networks that only connect when there is data to send. For example, in many retail environments, the remote store front sites only need to exchange data at the end of the day to update inventory and report the day's sales.
The other common type of network that uses only dialup connections involve sites that are in separate buildings, but within the same local dialing area. In this case, if the telephone company doesn't charge a usage fee, a pure dialup network can be a very cost-effective way of delivering low bandwidth WAN services.
Three technologies are commonly used for dialup links: standard analog telephone lines with asynchronous modems, switched 56Kbps synchronous digital service (sometimes called Centrex), and ISDN.
Standard analog telephone lines with asynchronous modems are a reasonably effective dial backup technology, and they have the great advantage of being nearly ubiquitous: in regions where you can get no other network services, you can often get an analog telephone line. Further, most Cisco routers have an AUX port that supports an analog modem connection.
But this option has some important drawbacks. The first is that there are no guarantees about how much bandwidth you will get. Many analog modems are rated to speeds up to 56Kbps, but in practice you will rarely get this much throughput. It is more typical to see a practical bandwidth of between 9.6 and 44Kbps with asynchronous modems.
The second important problem with voice grade telephone lines is that they are susceptible to electrical noise, which can cause dropped packets and sometimes even dropped calls.
Switched 56Kbps digital service, which also goes by the brand name Centrex in some areas, is a synchronous digital dialup technology. We recommend using this in regions that don't offer ISDN because it offers greater bandwidth and reliability than voice grade analog service. However, the number of local telephone companies that can offer switched 56Kbps but not ISDN is rapidly decreasing.
To use this technology, you need a synchronous serial port on your router, and an external Data Unit (DU), or synchronous modem.
ISDN (Integrated Services Digital Network) is usually the best way to go for dialup networking. It has the highest bandwidth and the greatest reliability. And, when using ISDN with Cisco routers, you have the distinct advantage of being able to use built-in ISDN terminal adapters and Network Termination Type 1 (NT1) units, which reduces both the complexity and the costs of implementation and maintenance.
ISDN circuits come in two basic varieties called Basic Rate Interface (BRI) and Primary Rate Interface (PRI). A BRI circuit supports two 64Kbps B-channels and a 16Kbps D-channel that handles the signaling for the two B-channels. A PRI circuit, on the other hand, uses a single 64Kbps D-channel to support the signaling for 23 (if delivered through a T1 circuit) or 30 (for an E1 circuit) B-channels. Many network vendors will also sell PRI services on fraction T1 or E1 circuits, allowing smaller numbers of B-channels.
The D-channel is not usually used for user data, but Cisco routers allow you to bind the two B-channels together for a net 128Kbps link using the PPP multilink feature. Unlike analog modems, each of these channels operates at full-duplex, so you can send and receive simultaneously at the full channel speed.
It is possible to use the D-channel of a PRI circuit for user data, but only if the carrier has not configured this channel to manage the B-channels. In situations where you have multiple PRI circuits, it is possible to control all of the B-channels from the D-channel of the first PRI circuit, leaving the D-channels of the other circuits available for data. The advantages of doing this are slight, however.
Many organizations use BRI interfaces for remote branch devices, and PRI interfaces for central dialup circuits. This way you can save on physical ports by having many branches dial into a single central PRI circuit. By default, a PRI circuit can accept calls from remote ISDN circuits. ISDN circuits can also terminate calls from Centrex or switched 56Kbps type circuits without requiring any special hardware. Further, Cisco has analog modem cards for several routers such as the AS5x00 and 3600 series. These allow you to terminate analog calls from remote devices on the same PRI circuit. This is an extremely useful option because you can then configure all of your remote devices to dial to the same central ISDN PRI telephone number.
BRI interfaces come in two main varieties, called "S/T" and "U." Usually a BRI circuit is delivered and terminated on a U interface, which is a two-wire digital telephone line. The U interface connects to an NT1, which converts the U interface signaling to S/T interface signaling. The S/T interface then connects to a Terminal Adapter device, which allows you to connect the ISDN circuit to your equipment. Both S/T and U interfaces use standard RJ-45 cables.
Cisco allows you to eliminate some or all of these pieces of equipment, though, by offering a variety of ISDN hardware options. Many access routers come with an optional on-board Terminal Adapter, or can take an ISDN module with this functionality. The BRI interface is labeled "S/T" to indicate when the router has an on-board terminal adapter. You can connect this port to an external NT1 device, which in turn connects to the telephone company's circuit.
Cisco also has a variety of BRI modules that include an on-board NT1. These also use an RJ-45 connector, but they are labeled "U" to indicate that you should connect directly to the ISDN circuit. We generally prefer to implement ISDN on routers with on-board NT1 units because it simplifies implementation.
If you want to take full advantage of ISDN features, the router must at least have an on-board Terminal Adapter.
Many network engineers make the mistake of either under or overestimating how many dial backup lines they need to provide at their central site. In a hub-and-spoke WAN, you can easily estimate how many dialup lines you will need at the central site based on the probability failure for a branch's primary circuit.
The most common failure mode in any WAN is the so-called "last mile" failure, which means that the local loop circuit between the remote site and the WAN provider's Central Office (CO) breaks for some reason. The break could be due to a fiber cut, cross-connection problem, or (more common than anybody would like) human error. The provider will usually keep statistics on these problems, which they will use to define their Service Level Agreement (SLA) for each type of circuit.
The SLA effectively reflects a probability of a circuit failure. If, for example, your remote sites have a 99.9% SLA, this means that there is a 0.1% probability of failure. So, if you have a network with N circuits, each of which has the same probability of failure, P, you can use the following formula to calculate the probability of k simultaneous failures:
The symbol "!" is a standard shorthand notation for the factorial function:
So, for a WAN SLA of 99.9%, which is on the poor side (but typical), P is 0.1% (100% - 99.9%). If you have a hub-and-spoke WAN with N=100 circuits, the probability of there being a single circuit down is:
So roughly 10% of the time, you can expect to have one circuit down. Similarly, the probabilities of there being two or more simultaneous failures are given by:
It's clear from this that the probability of 10 simultaneous failures is very small indeed. But just looking at probabilities can be deceptive because all of the numbers look small. We recommend multiplying these probabilities by the number of minutes in a year to get a better idea of how likely these failure scenarios actually are.
The probability of there being a single circuit failure is 10%, or 36.5 days per year. The probability of two simultaneous failures is 0.5%, which is roughly 44 hours per year. The probability of three simultaneous failure is .02%, or 105 minutes per year. And the probability of four simultaneous failures is .00038%, which is about two minutes per year.
So these are all things that you can expect to see happen at least once in the expected several year life span of this WAN. But the probability of 10 simultaneous failures is so small that you would expect it to happen roughly 5 x 10-10 seconds per year. Looking at this another way, if this failure condition lasted for one second, you would expect it to happen about once every billion years. Those are odds that most of us could live with.
By doing this sort of analysis, you can tell that having three dial backup circuits would probably come in handy at least once a year, and you might even need as many as four. But you're not likely to ever need 10.
However, it's important to bear in mind that this analysis assumes that these failures are not correlated. Depending on how your WAN provider implements your circuits, a single failure could affect several branches. So it is usually a good idea to apply a safety rule and double the number of circuits that this analysis suggests you will need. In this case, you probably need 4 circuits�but if you have 8 or 10, you should be more than safe.
 |
|
|
| Top |