Testing & Results

Testing Methodology

To test the ASUS PCE-AC66 PCIe Adapter I used two similar test applications, both of which are specifically designed for testing network throughput. The first one was Passmark Performance v7.0 Advanced Network Test. The second one was TamoSoft Throughput Test v1.0, Build 19. These tests measure throughput between two PCs connected through a router, switch, or a wireless connection. Normally, the networking hardware is considered the device that is under test. If you have a known baseline for several components in the network, say two identical network adapters and a switch, you can also test the performance of other devices in the communication chain, such as wireless adapters or network interface cards (NICs). In order for this test to work one PC must be set up as ‘Client’ and the other must be set up as the ‘Server’. Each test was run at least five times with the highest and lowest result omitted and the remaining results averaged to give a final result.

During earlier testing of GbE and 10GbE network switches, I’ve already established a baseline benchmark speed for these two workstations, Test System #1 and #2, below. Both of them are Quad-Core Intel systems, and can easily sustain Gigabit transfer speeds and higher, in wired mode. In the past, I used all default values for this benchmark, as shown in the screenshot below, but I discovered in this round of testing that certain combinations required a longer settling time before the true steady-state throughput was revealed. I’ll show an example of that, later. The user interface screens for the Passmark Performance test are the same at both ends, with different elements grayed out, depending on whether you are sitting at the Client computer or on the Server side. The TamoSoft application has different screens for the Client and Server sides, and only the client side displays the throughput graph that I’m showing here.

This benchmark eliminates most of the variables involved in network speed testing, but not all. The PCs themselves can introduce spurious issues, such as hardware bandwidth limitations, resource conflicts, wait states, and buffer inconsistencies. In some cases, the networking hardware is having issues communicating with other networking gear. The following chart shows why you have to dig a little deeper than just looking at the Average Transmission rate that is displayed on the main screen. For some reason, at the start of this test the two systems were having a hard time establishing a rapport, or the benchmarking tool is having trouble tracking the signal flow. The average value (shown in yellow) doesn’t really reflect the true capability of the network until things start to settle down around the two minute mark. The default time period that’s set in the software for this benchmark is 20 seconds (a looooong time in network transmissions), and the calculated average result at that point is more than 50% higher than the real number. I maxed out the test period to 200 seconds, and was finally able to see some convergence on the data rate after the first minute or so. BTW, the chart below is from a UDP test, where an awful lot of data was NOT making it across; more than 300Mbps was being sent out, and less than 20Mbps was being received and accepted. NOT a good result, as you’ll see….

For the second set of tests, I used an application that I discovered when trying to get the measure of some 10GbE networking gear that was on the test bed. Many of the current test products had difficulty measuring the full performance capabilities of a 10GbE network, but I did find a couple of applications that showed promise for normal test speeds. TamoSoft Throughput Test v1.0, Build 19 was the best of the bunch, especially in its ease of use and real-time graphical display of throughput. The graphs I show from Passmark Performance v7.0 Advanced Network Test are only available for viewing after the test is complete. Plus, there is a 200 second limit for the Passmark software, and I was able to run the TamoSoft test for as long as I wanted. Oh, and did I mention that you can run the TCP/IP and UDP tests at the same time? You can specify that only TCP data be used for the test, and I did run several tests to confirm that there were no differences, when running the UDP tests alongside the TCP one. Results for both upload and download are presented, and I ended up using the two download averages for reporting purposes. In the run below, that means a TCP result of 115.18 Mbps and a UDP result of 421.11 Mbps would be counted as the results. The packet loss (19.1 %) and Round-trip time (2.0 ms) are shown as well, but I didn’t report that. If the readers really want that data, let me know in the comments section and I’ll see about including it, somehow. TamoSoft has a bunch of additional test applications that are more comprehensive and are available for purchase, but this one is free for downloading.

Test System