Extreme’s S400, Part 2

In our last report, we talked about the Extreme Summit’s 400 Series 10G switch, which passed its performance tests with flying colors.

We also tested the Extreme S400’s performance in gigabit-only tests. While generally good, the results didn’t match the flawless levels we saw in the 10G Ethernet events.

Throughput was equivalent to 92% of line rate or better in all the tests we ran using standard Ethernet frame lengths. We observed a slight difference depending on whether we used random or non-random MAC addresses; on production networks, random addresses are far more common. This won’t necessarily harm application performance, since few (if any) production networks have sustained utilization over 90%.

Handling jumbo frames was another story. When we offered the S400 repeated iterations of 9,000-byte frames with a random pattern of MAC addresses, there was no rate at which the device would forward traffic without at least some loss. Throughput rose to the equivalent of 81% of line rate when we repeated the test using non-random MAC addresses. Again, random MAC addresses are the rule on production networks.

More seriously, the S400 dropped frames of any length after we ran repeated iterations of the jumbo tests with random addresses. Rebooting the switch cleared the problem.

Extreme acknowledged the issue, saying it is working on a fix.

It’s important to note that these results are reproducible only in controlled lab conditions. Those conditions – repeated 60-second blasts of traffic at or near line rate on 48 ports in a fully meshed pattern – are highly unlikely to occur on production networks. This is strictly a matter of specsmanship.

Even with the jumbo and MAC address issues we saw, delay and jitter remained remarkably low. The worst-case delay number of close to 200 microseconds with jumbo frames with random MAC addresses is nowhere near high enough to have an impact on application performance.

Jitter was also remarkably low, which augurs well for VoIP and video applications. With standard-sized frames and random MAC addresses, maximum jitter was around 3 microsec – at least an order of magnitude below the point where voice or video applications would suffer.

For the full report, go to https://www.nwfusion.com/reviews/2004/090604rev.html