Of Wires and Routers

If you ever wonder how your network packets get from hither to yon, here is a peek inside the UIC campus network, the ADN-ii. This article won't make you a network designer, but it briefly describes how we make network services both available and affordable.

The whole point of a computer network, much like a phone network, is to allow any two nodes (computers or people) to exchange information whenever they need to. One minor difference is that the data flow by voice doesn't change (even if you talk fast), whereas computers' appetites for information flow increases when their owners try to move ever-larger files faster and faster.

An important similarity, however, is that the number of potential connections is the square of the number of nodes. If one actually installed a physical connection for each possible connection, the cost of a large network would be astronomical.

• The ADN-ii
• What Do We Need to Keep on Growing?

A map of the ADN-ii network is shown in the centerfold of this issue. The most heavily shared piece of hardware on the ADN-ii is its fiber-optic backbone, which uses FDDI (Fiber Distributed Data Interface) technology. Although it may not look like it on the map, our FDDI backbone is topologically a ring. FDDI is fast -- 100 Mb/sec, and, in contrast to ethernet, the full bandwidth is achievable in practice. This is important, because just about every packet moving from one building to another (or off the campus) must travel over this part of the network.

Every so often along the FDDI ring, there's a router -- a highly specialized and very fast computer that is connected to the backbone on one side, and to several subnets on the other. Each router looks inside every data packet that comes to it on the backbone. It checks to see where the packet is going, figures out the best route, and sends it on its way.

At UIC, subnets -- the smaller networks that individual desktop machines are actually connected to -- are usually limited to a particular building or even to a particular floor in a building. Think of a subnet as your own cul-de-sac shared with other people on your floor or in your building; the FDDI ring is the freeway, and the router is a combination of on-ramp and traffic cop. The different subnets on a router don't interfere with each other, but you do share your subnet with everyone else on it.

At UIC, our subnets use ethernet technology, because it is much cheaper than FDDI. Ethernet provides a peak speed of 10 Mb/sec, but in practice, rarely gets above 4Mb/sec. For most purposes, such as word processing, some graphics, and medium-sized file transfer, this is just fine provided that the bandwidth is there when you need it. As long as your packet traffic is unsteady (usually the case), and there aren't too many people actively using your subnet, everyone on the subnet will work around each other, and everything will be fine.

Although topologically the ethernet subnet is a line into which everyone taps, in reality the entire subnet is contained in "hub", a box in a phone closet; the taps are extra pairs of telephone wires running through the walls of your building, leading to an ethernet card in your PC. (Actually, it's possible to have several hubs in a single subnet. This is necessary when a connection is beyond a certain physical length limit for ethernet cable; a hub is then needed as a "repeater" to extend the length.)

Another choke point is our connection off campus. Right now, we have two T1 telephone lines to the Internet, which give us a total of 3 Mb/sec for the entire campus. That's not enough for the high-end research and graphics currently being done on our campus, and also not enough to meet the needs of the campus as a whole, growing at our present rate. We're upgrading our external connection to an OC3 line (155 Mb/sec); we will be one of the first universities in the country to provide this much bandwidth to campus. Part of this bandwidth will go to our Internet connection, and part to connect us to the vBNS (very high-speed Backbone Network Service, the National Science Foundation's soon-to-come special supercomputer network) and other specialized networks.

An important consideration is that big, complicated things (like networks) break more often than small simple things, just because they have more parts. Improved ways of monitoring and fixing the network are essential if we are to regard the network like the phone system -- something that pretty much works all the time. We currently use SNMP (Simple Network Monitor Protocol) systems to monitor the status of routers from a central point. We'll be expanding this system to include more components and to give us a better picture of the nature of any outage.

As the network grows, so do the demands we make of it. As we use Web browsers like Netscape and Mosaic more heavily, as we begin using X Windows, animations, and interactive distance learning, even peak ethernet speeds won't be enough. So we are evaluating new technologies, including switched ethernet and ATM (Asynchronous Transfer Method), for possible use for future expansion.

The network, just like the PC on your desk, is not something that the campus can buy once and forget. We need to plan for regular upgrades, to keep up with a growing number of people using more demanding applications. One can't stand still in this business, but it is fun to keep jogging along.

Comments are appreciated; send them to
Bob Goldstein, bobg@uic.edu