Wireless Performance - more
If we look at the IxChariot plots, we can see that peak throughputs during the one minute test runs are actually higher. Figure 11 shows downlink, where a 148 Mbps maximum throughput was recorded. Large periodic downward spikes and high variation both help to push down the average throughput number.
Figure 11: Buffalo WZR-D1800H wireless throughput, 5 GHz, 80 MHz mode, downlink
For uplink, shown in Figure 12, peak throughput is 151 Mbps. Many more big ol' throughput dropouts, however, push down the average value significantly, however.
Figure 12: Buffalo WZR-D1800H wireless throughput, 5 GHz, 80 MHz mode, uplink
To be fair, however, we also need to check the three-stream N plots for the EA3500 and RT-N66U. Checking the EA3500's 5 GHz, 40 MHz mode, three-stream downlink plot shows only a 119 Mbps peak throughput, no match for the 148 Mbps shown in Figure 11. But checking the RT-N66U's 5 GHz, 40 MHz mode, three-stream uplink plot reveals a peak throughput of 170 Mbps, which is 20 Mbps higher than the D1800H's 151 Mbps peak shown in Figure 12.!
My takeaway from all this is that, for a single data stream, draft 11ac didn't produce significantly higher average throughput than you can get from three-stream 802.11n. And keep in mind that three stream N is doing this while eating up half as many 5 GHz channels!
How High Can You Go?
But that's not the whole story. I know from previous testing that running multiple streams will produce higher total throughput than you can get from one client. That's why I added the up/down benchmarks to the Charts. Figure 13 shows the up/down chart for 5 GHz, 40 MHz mode, three-stream N.
Figure 13: Three stream N product up/downlink throughput - 5 GHz, 40 MHz mode
If you compare Figure 13 to Figures 9 and 10, you see that throughput gain from using multiple pairs varies greatly from product to product.
So I ran the up/down test on the Buffalo draft 11ac pair and got 170 Mbps! This was a pretty big jump from the separate up and downlink runs, so I decided to keep going and add on more pairs. It turns out that I could add a lot of pairs and still get higher total throughput. The plot in Figure 14 shows that seven up/downlink pairs was as high as I could go before significant throughput gain topped out at around 440 Mbps of total throughput.
Figure 14: Buffalo Draft 802.11ac - Throughput vs. Traffic Pairs
This is impressive! But could I do the same thing with three-stream 802.11n? To find out, I took a Cisco EA3500, paired it with my standard Intel 6300 three stream test client and started testing. Figure 15 compares the test runs and shows not only a flatter curve for three-stream N, but also that it took only four pairs for gains to stop (actually more like three pairs).
Figure 15: Draft 11ac vs. Three Stream 802.11n throughput vs. traffic pairs
In the end, I was only able to get 165 Mbps of total throughput from three-stream N. So using the Buffalo draft 11ac router and bridge produced 266% higher total throughput than three-stream N.
The conclusion from this experiment is that draft 11ac is more likely to benefit busy multi-user WLANs than to help ensure more reliable wireless HD streaming. It's possible that folks who want to run multiple HD streams might benefit, but remember that per-stream throughput drops as more streams are added.
Figure 16 shows 80 Mbps down and 90 Mbps up average per-stream throughput vs. the 125 / 117 Mbps down / up throughput shown back in Figures 11 and 12. The big throughput dropouts won't be kind to HD streams either.
Figure 16: Draft 11ac - one pair
Adding two more streams for a total of four, drops per stream throughput to around 70 and 80 Mbps for down and uplink, respectively. Still not too bad, and certainly higher than you get with even three-stream N. But the throughput is certainly moving in the wrong direction for reliable HD streaming.
Figure 17: Draft 11ac - two pair
I think I've ridden the "How Fast Will Draft 802.11ac Go?" horse as far as it will go. I have to say I am impressed with draft 802.11ac's ability to provide unprecedented aggregate throughput when handling multiple data streams. Now I can move on to Part 2, where we'll look at throughput vs. range for as many modes as I can stand testing.