Benchmark Results: PerformanceTest, 2.4 GHZ
Now we can peek under the hood and see what’s really going on with these adapter connections. There are a lot of different ways to compare this data, so bear with us as we try to illustrate a few things.
Curious, huh? In same-room conditions, looking at 2.4 GHz TCP traffic, the Killer adapter easily overleaps its competitors, but once we add in obstructions and distance, Intel leaps ahead. Now, keep the decimal places in mind and watch that x-axis scale. By the time we reach Location 3, all three adapters are barely passing bits. Intel is the top dog in a race of two-legged canines here. Also note that, predictably, 16 KB block performance is usually faster than 4 KB, regardless of distance.
We don’t know why the Killer 1103 fell down on our Location 2 UDP test, but check out that Location 3 result. Maybe everybody else in the neighborhood turned off their routers at the same time for three minutes. Whatever happened, that’s some performance spike for Qualcomm. Unless the Killer component has some secret sauce for how it handles UDP traffic (doubtful given the Location 2 results), we’re a bit stumped here.
But this raises an important point about the nature of wireless testing and how immensely variable it can be at any given time.
A first look at Cisco’s 2.4 GHz TCP throughput illustrates very well the increasing havoc that distance and obstructions can wreak on throughput. Keep in mind that an ideal chart would look like a flat line at the top of the graph. Location 1 is decent in that we get a fairly even average (yellow line) with no major spikes or troughs. Location 2 gives more cause for worry, as performance steadily declines starting about one-third into the test. This might be due to ambient conditions and have little to do with Cisco’s design. Location 3 shows serious pain, in which the connection is only able to stay alive through occasional gasps. As you can see, barring those eight or nine blips of data transfer, Cisco’s Location 3 throughput is essentially dead.
When we compare different adapters in the same location, we achieve an even deeper level of insight. Cisco swings up from a lower performance zone into one of higher performance, although we still see multiple sharp troughs. Intel maintains a flat average, but look at the range of its throughput swings compared to Qualcomm. In this particular instance, Intel would be the better choice for, say, file transfers because it’s average throughput is considerably higher than the Killer part. However, if we were streaming a 5 MB/s video feed, we would definitely opt for the Qualcomm because, unlike the Intel 6300, it maintains a smooth, steady rate well above our throughput needs without suffering those momentary troughs that cause dropped frames.