Network Computingwww.networkcomputing.com

The technology is also a great way to replace multiple gigabit trunks between switches. Trunking makes it possible to connect multiple 1-Gbps ports together as one virtual connection, because the traffic is equally balanced among the connections. The disadvantage of this approach is that it requires fiber for each connection, and it can be difficult to troubleshoot. Aside from using less fiber, one 10-Gbps connection provides a cleaner solution.

During the year, Enterasys will be rolling out all four versions of the physical interfaces that will be supported by 802.3ae, which will include longer-range options running over single-mode fiber. For now, however, you will need Matrix E1 Optical Access Switches on both ends of your 10-Gbps link. Standards are important, but if you require more bandwidth now, the E1 will provide it until the standards details are worked out.

The E1 was designed as a Layer 2 switch, with no pretense of being anything else. The final code will include support for capabilities you would hope to find in a Layer 2 switch, such as 802.1q and 802.1p trunking, multicast support, VLAN (virtual LAN) support and port monitoring. I was disappointed that Enterasys wouldn't let me examine these features in the beta version of E1 I tested, even with the greater flexibility Network Computing gives beta products for Sneak Previews (as compared with feature reviews). By the time you read this, however, Enterasys will be delivering its production product, so you'll be able to test the features for yourself. The company also will be adding 10 Gigabit interfaces to its Layer 3 switches.

Maximum Bandwidth

The E1 I received from Enterasys came with 12 fiber gigabit ports and a modular 10-Gbps uplink. The gigabit ports all had MT-RJ interfaces, which are each about the size of an RJ-45 connector. Four of the 12 interfaces were GBICs (Gigabit Interface Converters) -- hot-swappable adapters that let you easily change from single-mode to multimode fiber interfaces. The GBICs also support MT-RJ connections, making it possible to fit all 12 interfaces in a small form factor, which comes in at 1.5 Us or 2.6 inches high.

I was able to use the E1's telnet, Web and serial-port management capabilities to do basic tasks, such as configuring an IP address for the switch. One thing I can say for sure is that the E1 can move frames -- after all, that's its main purpose. I hammered the product with a barrage of tests using a pair of Spirent Communications SmartBits 6000B testers (see "How I Tested 10 Gigabit," below). No matter what I tried, however, I was unable to get the E1 to drop a frame. The switch exhibited 20 gigabits of unadulterated wire-speed performance.

I used a variety of frame sizes for testing. The minimum-size frames, 64 bytes, are the most difficult to forward without frame loss -- the smaller the frame, the more frames that are transmitted per second. When running frames through a switch, the Ethernet header of every frame needs to be inspected, and the bridge table has to be examined to determine the port from which to send the frame.

It's possible for a 10-Gbps connection to handle nearly 15 million 64-byte frames per second in each direction, for a total of almost 30 million frames per second. Therefore, the E1 had no problem handling the nearly 30 million frames per second it had to process when I blasted it with 64-byte frames.

Whether or not you are ready to make the jump to 10 Gigabit Ethernet on your network, the Matrix E1 Optical Access Switch is ready to produce exactly what Enterasys designed it for: raw bandwidth.

• Full-Mesh Throughput Test. For this test, I transmitted different size frames at 100 percent utilization. With each port transmitting at 10 Gbps, the total utilization in each direction was 10 gigabits. I used a full-mesh test setup so that every 1-Gbps port divided its traffic evenly among all the corresponding ports on the other end. This made the switch work a little harder than sending to just one port. The result was that even with smaller frame sizes, which caused the switch to process more frames per second, the switch didn't drop a single frame. This is known as true wire-speed throughput.

• Latency Test. For this test, each gigabit SmartBits port generated frames and measured the time it took for them to arrive at the opposite SmartBits port. Multiple frames were measured, and I took the worst case, or the frame that exhibited the highest latency, for each test. Even the worst case showed such low levels of latency that it would not be significant, even for sensitive applications like VoIP (voice over IP).