Results: PerformanceTest 7, Same-Room
PerformanceTest 7’s network test is similar to IxChariot in some ways, and it provides a graphically friendly way to cross-check other benchmark results. Here, in our 5.0 GHz TCP same-room test, we see Asus pull back a bit, even slipping in behind AirLive. Buffalo, Linksys, and Netgear all perform in the 165 to 180 Mb/s range, which, as a general average, meshes well with what we saw in our 2 GB transfer tests. In fact, Asus is the only player here that sees a significant difference between the two benchmarks.
In moving to UDP, throughput suddenly skyrockets, then crashes into some sort of bottleneck. Netgear's R6300 is the only router to not pass the 600 Mb/s mark.
To double-check our assumption and find out why traffic was maxing out, we went to David Wren, the creator of PerformanceTest.
“My guess at what is happening is that the device driver is just accepting an unlimited amount of data, then what can’t be sent with the available bandwidth gets thrown away,” he replied. “UDP was designed for tasks like streaming video and VoIP. So it would be like trying to push a super high definition video across the link, but on the receiving side you find that five out of every six video frames didn’t make it. But from the point of view of the sending application, everything is sent. In real life situations UDP is not used for transmitting as much data as possible as fast as possible. It is used when data needs to arrive on time, meaning that lost data isn’t worth recovering (or retransmitting) because you know that new data, e.g., the next frame of the streaming video, will arrive soon enough. You might find the CPU or PCI bus maxed out at 600 Mb/s, preventing even more data from not being transmitted.”
When we inquired as to why IxChariot’s UDP data might be so much slower than PerformanceTest 7’s (as you’ll see in a bit), Wren was quick to point out that he has never used or investigated IxChariot. However, he suggested this:
“From what I have read here, it looks like [Ixia] have implemented (by hand) their own version of TCP (with ACKs, sliding windows, and retransmissions) on top of UDP. Quote: ‘...This datagram protocol is a subset of the functionality TCP provides to ensure that data is received reliably....’ Which doesn’t really make sense to me, as no one in real life would do this. If you want a reliable connection, then you use TCP. If you want a lossy connection, then you use UDP. If I understand their paper, they are really measuring two different versions of TCP, the full Winsock implementation and their own custom coded TCP-like protocol.”
Switching to 2.4 GHz TCP, we again nose dive well below 802.11ac levels, just as in our 2 GB transfer tests. Asus easily trounces the field, with Buffalo coming in second and trailing by over 40 percent. So far, we’re having a hard time reconciling the 802.11n performance of these units, as we’ve seen prior-year models that perform better on 2.4 GHz for half the price.
In the 2.4 GHz UDP test, Netgear recovers into the 600+ Mb/s range, with overall scores running only slightly lower than in our 5 GHz pass.
