eWEEK content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.
In order to give Cisco’s Catalyst 4900M switch a run for its money, I took the unit to Ixia’s newly inaugurated iSimCity test and measurement center in Santa Clara, where I subjected the switch to Layer 2 and Layer 3 tests, alongside a set of power consumption tests.
I conducted performance tests on a 4900M unit that was outfitted with two WS-X4904-10GE half cards that each carried four 10 Gigabit Ethernet fiber ports. Taken together with the 4900M’s eight fixed 10GE ports, this brought my test unit to 16 ports of full line-rate 10GE capacity.
I generated test traffic with Ixia’s Optixia XM12 IP performance tester chassis with six LSM10GXM3-01 cards that each had three 10GE ports. I conducted all of the tests in store-and-forward mode, in which the frame size was subtracted from the latency calculation to single out latency introduced by the 4900M.
Our first set of tests measured the 4900M’s wire rate forwarding and switch latency with L2 traffic. I used Ixia’s “RFC2544-IPv6 Benchmark Latency” test, which is actually an IPv4 benchmark that Ixia recently updated to include IPv6 support. I tested only IPv4 traffic. I ran the 4900M as a Layer 2 switch at 100 percent utilization across all 16 of its 10GE ports, with frames sized at 64, 128, 256, 512, 1024, 1518 and 9216 bytes.
The results were impressive. Across all frame sizes, latency hovered between 2.5 and 2.6 microseconds. There was no appreciable difference in frame forwarding latency in any test run. All test loads were run for 5 minutes to ensure that any buffering or other fault conditions that could arise from sustained maximum performance would be shown. In all maximum utilization tests–in fact, for all performance tests that I ran–there were no frame handling errors reported by the Ixia test tools.
I ran the same set of tests at only 10 percent utilization to check for inconsistencies between full utilization and a somewhat more normal utilization rate. For each frame size, our tests yielded almost identical latency results–forwarding latency remained flat at around 2.6 microseconds across all frame sizes.
I also ran the Ixia IPv6 Benchmark in Layer 3 routing mode at both 100 percent and 10 percent utilization with similar results. Latency stayed at approximately 2.6 microseconds with no frame transmit or receive errors across all frame sizes. These tests I also ran for 5 minutes to check for problems at sustained maximum utilization rates.
During the L2 switch testing I measured the total power consumption using a Watts Up Pro watt meter from Electronic Educational Devices. At full throttle, the 4900M consumed 242 watts, which is about 15 watts per port.
The 4900M has dual power supplies and a single fan tray. The power supplies are hot-swappable and during full load testing I pulled one power supply for 20 minutes of a 45 minute test. The 4900M continued processing without exhibiting any errors or noticeable increase in fan activity. When the power supply was reconnected, the 4900M continued functioning with no change in switch activity.
I did not pull the fan tray. The manual specifies that the 4900M can run up to 15 minutes without the fan tray, which should be enough time for a technician to perform a field replacement.