IxChariot TCP And UDP Multipair
We saved the best scenario for last. Most homes don’t have exceptional problems with interference or distance, but we are, on the whole, gradually increasing the number of clients connecting to our LANs. All it takes is a couple of kids with tablets, and you’ve got potentially three concurrent media streams plus a couple of guests pulling data to their phones plus whoever is surfing the Web...and so on. The idea of having six or more network clients active in a mainstream home is no longer crazy. Hopefully, at least some of these will be wired, but odds are that at least some will be wireless, and your router has to be up to the job if you want everyone to stay happy.
As you might imagine, running repeatable content to two, four and eight clients simultaneous for all of these tests would be questionably accurate and unduly laborious. Fortunately, we can find a close proxy in IxChariot’s multipair simulations, which also allows us to isolate TCP and UDP traffic.
Keep in mind that the baseline tests use only one pair—data flowing to one client and back from that client. In these tests, we’re simulating multiple clients. If the router can handle the extra load, we should see performance scale up as more pairs get added. For an analogy, imagine you’re trying to figure out how many people can go up a narrow escalator in one minute. If you send them one at a time (one client), you’ll get your baseline. If you gather your crowd into pairs and have them hold hands (two clients), they might get a little crowded standing side-by-side, but overall you’ll get more people to the top in a minute. When you send groups four at a time, you might see more incremental improvement, or you might have your people jostling for space so much that they have trouble getting processed. By the time you reach groups of eight, you better have someone at the escalator helping to organize people as they get on, otherwise you’ll have a mess if not a fistfight.
We’ll provide the baseline again as a reference. Keep in mind that we’re only interested in the average throughput results.
We clearly see Amped’s RTA15 improve as we add clients, though the gains are modest. The difference in total throughput from one client to eight is only 13%, and the needle hardly budges in going from four clients to eight. Asus, sad to say, clearly buckles under even two clients. It tries to recover with four but still doesn’t even come close to its single-client throughput. The same holds true for the R6300. Only the Nighthawk takes a deep breath and keeps lifting as more plates keep going on the barbell. Admittedly, we don’t see much improvement from four to eight, but at least it doesn’t buckle under the load.
If you’re curious about what’s happening under the hood, check this out these eight-pair charts from the Asus (left) and Nighthawk:
In general, we saw routers spend the first 10 to 20 seconds of this test exhibiting very divergent performance across their eight pairs. It was as if the routers were organizing streams and allocating resources on the fly. Once they got things figured out, streams performed in lockstep. But as you can see, lockstep or not, Asus struggles to maintain a steady data flow over the test’s run. In comparison, the Nighthawk’s graph almost looks like a mountain lake at sunrise, smooth as glass and ready for some serious skiing.
When we switch from 5 to 2.4GHz, still sticking with TCP traffic only, we get a different story. Amped and Asus both show no ability to scale beyond one client. The R6300, strangely, finds its legs and doubles its throughput when moving from one to two clients, but then won’t budge any higher. The Nighthawk sees a little lift when stepping up from one client, but then, like the R6300, quickly hits its ceiling. Amped actually has the higher total throughput under load here, but we had to give the win to Netgear since the RTA15 lost 20% of its performance in the shift.
For media streaming, we care most about UDP performance. We limited our testing to two and four pairs because none of the routers could do eight UDP pairs without either IxChariot puking up warnings or our results showing an extraordinary amount of lost data. How much is extraordinary? While there is no official ratio, we figured that anything above 25% constituted so much loss that throughput results would be somewhat meaningless. For example, a two-pair UDP test should, given our parameters, transmit 2000 timing records. If the results only show 1000 records received, then that represents a 50% data loss. You’re not going to get much enjoyment out of half of a data stream.
But again, the Nighthawk is clearly the best bet. All routers do manage to increase UDP performance on 5GHz as we increased client load, but all except the Nighthawk reveal significant error strain with even four clients.
Again, we walk away disappointed with 2.4GHz scaling with UDP traffic. While the Nighthawk remains the only router free from errors and marked data loss, not even that unit could eke out a performance gain beyond one client. We hand Amped the win since it's the only router to show any positive scaling, however slight, although we’d probably still pick the Nighthawk for this application in order to have higher-quality streams.