RSS
  The Road To Fast Ethernet Networks In addition, look out for the propagation distance of Fast Ethernet--the entire distance between your data center and your farthest node cannot exceed 205 meters without going to fiber-optic connections. Furthermore, there can only be 5 meters between two cascaded hubs, making traditional cascaded networks virtually impossible. To solve this one-two punch, we introduced a few Fast Ethernet switches into our network. Switching splits the Fast Ethernet collision domain into manageable chunks and adds powerful proactive management tools such as Remote Monitoring (RMON). We ran seven of our runs back to a 3Com SuperStack II Switch 3000 TX, and the other six to an Intel Express Switching Hub. Each of the hubs was connected to a port on the Catalyst 5000 in our network closet. For a more scala ble solution, we could have put a 12-port switching board into the Catalyst. Using the two ports on our Catalyst 5000 saved about $12,000 in our implementation, but reduced our total available bandwidth by about 85 percent (from 1,400-Mbps to 200-Mbps maximum throughput). Another way to add distance to your Fast Ethernet network is to use two-port switches to extend your network diameter. Several vendors offer units that let you bridge networks from 10 Mbps to 100 Mbps and even 100 Mbps to 100 Mbps. These cost around $1,000, and allow you another 200 meter hop on your network. The drawback to these units is that they may not be able to keep up with wire-speed traffic. Bringing It All Together At the Computer-Aided Engineering Center (CAE), the servers are located in a single building, and the separate facilities are connected via an FDDI ring. The backbone is distributed campuswide with FDDI-attached servers. To get our new Fast Ethernet network onto the ring, we would have had to buy a Fa st Ethernet port for our Cisco router or use the two switch ports on our Catalyst 5000. We chose to use the existing switch ports on the Catalyst 5000, since our AGS+ router does not (and never will) support 100BASE-T uplinks . This had the unintentional effect of flattening our previously routed network. In fact, by adding 84 machines directly onto the ring, we ran out of IP addresses on that segment of our subnetted Class B network. We were able to circumvent the problem by expanding our ring's IP range to include 1,024 addresses, thanks to our Cisco routers. Note that not every router manufacturer allows for variable subnet masks on different segments. One possible solution is to use Ethernet virtual LANs (VLANs), but that requires switches that can communicate VLAN information between one another--a standard (802.1Q) for this is still several months down the road. Of course, buying one or more Fast Ethernet router ports also would have alleviated this problem, but would have greatly increased the cost of this project. In the lab, the workstations are clustered into groups of eight to 12 workstations. We used 100-Mbps hubs such as the 3Com SuperStack II Hub 100 TX, the Intel Express 100BASE-TX Stackable Hub, the AsanteFAST 100 Hub and the Dayna BlueStreak 100 Fast Ethernet hub to connect the workstations to the switches in the wiring closet. New Ethernet interfaces and a flattened network forced us to reconfigure our Dynamic Host Configuration Protocol (DHCP) server for the proper Media Access Control (MAC) and IP subnet information on the new network. After we were confident everything was ready to go, we physically moved the CAT5 cabling from the 10-Mbps hubs to the newly installed 100-Mbps hubs, freeing the old repeaters from the network. Everything appeared to go smoothly, but there were some issues that would come back to haunt us later. More Than Meets the Eye Although our CAT5 network was carefully planned and wired, we still encountered problems. The CAT5 specification is demandin g, not only in the walls, but also at the ends. Specifically, we were encountering an unusually large number of frame alignment and CRC errors on our Cisco Catalyst 5000. Our diagnostic equipment told us that our wiring was properl y constructed, but there were subtle flaws at the ends of some of our runs. Another curious phenomenon was that our workstations were occasionally losing connections to the application servers, much to the distress of our users. To try and sniff out the problem, we borrowed a MicroTest PentaScanner 350 and certified each of our CAT5 runs. We were surprised to find several failed runs, mostly due to inverted polarities or crosstalk at the connector. Recrimping several ends solved our problems, but a standard cable tester was unable to find the crosstalk problems, which the more sophisticated PentaScanner detected. The higher demands of Fast Ethernet may warrant the purchase of a higher-grade cable scanner--one capable of finding problems such as these.
|
|
|
by Thach Vo Updated Februayr 7, 1997 |
