| Previous | Table of Contents | Next |
Classic Ethernet runs at 10Mbps (megabits per second). That is, 10,000,000bps (bits per second) can be transmitted on the wire. Divide this by 8 bits per byte, and you get 1,250,000Bps.
Bits per second is indicated with a lowercase b, whereas bytes per second is indicated with an uppercase B. Therefore, 10 megabits per second would be 10Mbps, and 10 megabytes per second would be 10MBps.
That’s a lot. It means that about 1MB can be transmitted on the wire every second. Does this really happen? Not unless the Ethernet link is a point-to-point connection without other stations trying to share it. Even though 97 percent of the data is actual “payload,” every time you get a collision, it decreases the efficiency of your network and the whole blessed packet has to be retransmitted.
Are collisions a fact of life? You bet. Each time you add more than two Ethernet stations to a segment, you add to the potential for collisions. No big deal; even if you drop 30 percent of your packets due to collisions, you still see about .8MB per second, which is pretty fast.
However, in a highly populated—and thus highly trafficked—Ethernet segment, the collision rate might approach 80 percent. (Makes you wonder why you don’t just use a floppy disk.) On a segment such as this, problems might manifest themselves in various ways. Because some applications are timing sensitive, for example, if they don’t get their data quickly enough, you might see all kinds of errors. Therefore, it pays to take a gander at your physical documentation and count the hub ports.
The collision problem leads some Token-Ring proponents to smugly assert that this is why they use Token-Ring instead, because it scales much better and can truly provide the bandwidth it advertises. Don’t worry. Ethernet is fine. I’ll talk about the virtues and the problems of Broken-Ring—I mean Token-Ring—in the next hour. Meanwhile, the truth is that all network topologies have their problems—the key is to recognize them and to deal with them in a proactive way that causes you to lose the least amount of hair. (I’ll bash Ethernet a little bit during Hour 10, “Token-Ring Basics.” It’ll be fun, so stay tuned.)
Keeping track of how many workstations are connected to a segment becomes even more important with 100Mb (megabit) Ethernet. (We’re not even going to get into 1,000Mb Ethernet, or gig Ethernet, because it currently requires fiber-optic connections to achieve these speeds as well as professional installation and maintenance.)
Although 100Mb Ethernet works exactly like Ethernet—even down to the frame size and collisions—it runs 10 times faster. So, instead of about 1Mbps, we’re talking about 10Mbps. Presumably, the reason folks install 100Mb Ethernet is to actually get a high-speed network. A large collision domain (fancy talk for “lots of people on a shared segment”) can bring your throughput down substantially, so it makes sense to use switching gear to cut down the “party line babble” and run as close to 100Mb as you can.
For your servers, a full-duplex fast Ethernet connection might be the antidote to a slow connection. A full duplex connection to a switch means that you’re using both the transmit and the receive wires simultaneously; therefore, you double the speed. This means your 100Mb Ethernet pipeline becomes a 200Mb Ethernet pipeline. Wow!Half Duplex, the typical Ethernet or Token-Ring connection, is like when you use a radio. One person talks, then the other person talks back.
Full duplex is like using a telephone—both stations can talk at once. Because this is not a party line, and because computers (unlike you and I) can do lots of things at once, this does not cause collisions.
Just make sure your Ethernet driver as well as the switch you’re going to connect to both support full-duplex operation; the manuals should tell you how to enable this. Most cards and switches will be capable of performing at full duplex if you’ve purchased them recently, but it couldn’t hurt to check.
| Previous | Table of Contents | Next |