If all clients use the same standard, bandwidth sharing should be equal... theoretically. In the real world, differences in adapter design, host platform OS and architecture and RF environment result in somewhat less than ideal throughput sharing.
During the last round of experimentation, I ran a quick check using two NETGEAR A6200 AC1200 class USB adapters connected to a NETGEAR R7000 router and got the result shown in the IxChariot plot below.
Two AC1200 clients on AC1900 router
The host machines used had both different processors (Intel Core i5-4200 vs. Core i3-2310M) and OS (Win 8.1 64 bit vs. Win 7 Home Premium SP1 64 bit), but the same adapters. Two interesting observations from the experiment are:
- Bandwidth sharing was not equal
- Total bandwidth use did not exceed the 199 Mbps observed with a single AC867 client
The second observation is particularly interesting because the test router is AC1900 class and produced a peak downlink throughput of 409 Mbps with our AC1750 class standard client in the ideal conditions of our closed test chamber.
NETGEAR R7000 - 5 GHz downlink throughput profile
But inquiring minds also wanted to know what happens when different AC class clients are mixed. So let's take a look.
First, let's recap the products used. The test router for these experiments was again the NETGEAR R7000 "Nighthawk" [reviewed] loaded with V126.96.36.199_1.0.15 firmware. There was no particular reason for choosing this router other than I had it on hand.
I used three test clients with the router, representing AC1750, AC1200 and AC580 classes. These have maximum AC 5 GHz link rates of 1300, 867 and 433 Mbps, respectively. So they are identified in the plots as AC1300, AC867 and AC433.
- AC1300: Our standard AC test client composed of a Dell Optiplex 790 Small Form Factor (SFF) computer (Core i5-2400 @ 3.1 GHz / 4 GB RAM) running Win 7 Pro with an ASUS PCE-AC66 AC1750 PCIe adapter with Win 7 188.8.131.52 driver. The adapter uses a Broadcom BCM4360 chipset.
- AC867: Acer Aspire S7-392 running Win 8.1 64 bit with NETGEAR A6200 AC1200 class adapter [reviewed] with 184.108.40.206 utility and 220.127.116.11 driver. The adapter uses a Broadcom BCM43526 chipset. The Acer's internal Intel Wireless-N 7260 adapter was disabled.
- AC433: Lenovo X220i running Win 7 Home Premium SP1 64 bit with Linksys AE6000 AC580 class USB adapter with 18.104.22.168 driver. This adapter uses a Mediatek MT7610UN chipset.
The router and all three clients were set up in my test lab with all clients within 10 feet of the router. Tests were run with the router set to channel 153 using open-air testing and IxChariot's throughput.scr script with TCP/IP and test file sizes of 5,000,000 Bytes. No other networks were operating during the tests.
The Tests - Solo
As before, the three clients were run individually to establish their maximum throughput levels. The composite plot below shows the AC433 class Linksys adapter actually produced higher throughput than the AC867 class NETGEAR! The throughput jump 15 seconds into the AC1300 plot is an artifact of IxChariot's TCP/IP handling under high-throughput.
The AC1300 Mbps client produced 220 Mbps average throughput and 260 Mbps peak. This establishes the maximum downlink throughput available for this test setup. If the router can support all three clients at their maximums, it will need to provide over 530 Mbps of total bandwidth. It clearly can't do that. But will it produce the 409 Mbps peak we got in our chamber testing?
Composite plot of individual tests - downlink
For the sake of completeness, here's the composite plot of three individual uplink tests. This time, the AC867 class adapter did much better and the AC433, much worse. But average throughput for the AC1300 client was again around 250 Mbps.
Composite plot of individual tests - uplink
The Tests - Simultaneous
The next plot shows the three AC clients run simultaneously. The router didn't cough up the 530 Mbps required to support all three at their maximum bandwidths. Instead, throughput totaled only 246 Mbps with plenty of variation for all three streams as they competed for bandwidth. This is slightly below the 250 Mbps average produced by the AC1300 adapter running by itself.
Simultaneous tests - downlink
Things were a bit more well-behaved running uplink as the plot below shows. But again, total throughput does not exceed the maximum produced by the single highest-class (AC1300) client running alone.
Simultaneous tests - uplink
I also tried a downlink run, staggering the starts of each device by 40 seconds. The result below doesn't show the extreme battling for bandwidth produced by the simultaneous starts. But you definitely see an increase in the AC1300 client's throughput variation as the other clients come online.
Staggered tests - downlink
The Tests - Bandwidth Caps
With mixed-standard networks, we found bandwidth was shared more equally when bandwidth caps were applied to each client. So we tried this again, with similar results. The plot below shows each client's transmit rate limited to 50 Mbps. Couldn't get more equal sharing than that, eh?
Simultaneous tests - downlink- 50 Mbps rate cap
The Tests - Bandwidth Caps - more
With around 250 Mbps of bandwidth available, I should be able to set the caps at 80 Mbps and still have all three clients sharing nicely. So I naturally set the limit at 90 Mbps and got the results below. While none of the clients hit its cap, bandwidth was divided surprisingly equitably.
Simultaneous tests - downlink- 90 Mbps rate cap
To show this nice behavior has its limits, I raised the bandwidth caps to 100 Mbps and got the result below.
Simultaneous tests - downlink- 100 Mbps rate cap
Once again, in no case did total bandwidth exceed the 250 Mbps best case obtained with the AC1300 Mbps client running solo.
We shouldn't be surprised by these results. Channel capacity (bandwidth) is determined by the Modulation and Coding rate (MCS index) and number of spatial streams in use (see chart). In the case of our test router, the maximum usable throughput was measured at ~ 250 Mbps with a client capable of linking at the router's maximum 1300 Mbps 5 GHz rate. So even if we had all AC1300 (3x3) AC clients, they still would have only 250 Mbps of bandwidth to share.
So the answer to the question of how much throughput you can get from an AC router is: no more than the maximum supported by the combination of your router's class and its highest-class client. Adding more clients, even if they support the router's maximum rate, only divides the available throughput among them.