Does a wireless-N (802.11n) network have poor performance when in b/g "mixed" mode?

Question

It seems to be an "old" wives tale that, when using a wireless-N router in "mixed mode" to support legacy 802.11b or 802.11g devices, the performance of 802.11n clients will suffer.

---

Some places claim that when running in mixed-mode, all (some?) N clients run at G speeds. Others make the same claim, but say it only happens when a G client is connected.

Other places say N clients run faster, but still run about 30% slower than if the router were in N-only mode, even if there are no legacy B/G clients connected.

Still others claim there is no speed drop for N clients when running on a mixed-mode network. They say the only issue is that overall network throughput will be lower, because only one client can be transmitting at any time, so some of that transmission time must be shared with the legacy B/G clients running at lower speeds, reducing the overall throughput from what it would be if there were only N clients connected.

---

So, which is it? Will running in mixed-mode slow down my network, even if there are no B/G clients? If I'm running N, will having another client connected at B/G slow me down substantially vs. if they were running N?

Solution

From Beginners Guide To Networks :

Q: Does the mere presence of an 802.11B device slow down an otherwise all-G or all-N network?

A: YES This is of course already well known, although the exact implications are often misunderstood. The presence of an 802.11B device on an G or N network causes the newer devices to have to resort to some kludgy behavior to make sure that the B devices don't transmit when the G/N devices are using the airwaves, and to make sure that both the B and G/N devices can see things like beacon packets.

The exact impact on throughput is hard to estimate generally, but it will NOT "slow the whole network to 802.11B" as is often stated. There is however a significant slowdown imposed by the mere presence of a B device, even when it is not active. We (Slim) did some testing of this a couple years ago and found that usually the throughput between the G devices dropped by 30-50% (eg from 20Mbps to 10Mbps), but not nearly as low as the speed of a B-only network (5Mbps in the same environment). The theoretical maximum throughput on 802.11g is 23 Mbps without any B devices associated, and 14Mbps with.

Q: Will 802.11G devices slow down an all-N network?

A: NO, except insofar as the air-time that they take when active will be at the G throughput level as opposed to the N level. I.e. the devices still each communicate at their optimal rate in each time slice.

Unlike in the 802.11B backward compatibility mode, G devices do not impose any performance-degrading behavior on N devices in order for them to be backward compatible. 802.11g devices are able to recognize the 802.11n preamble, and they play nicely in terms of knowing when one or the other is trying to transmit. The preamble tells which modulation scheme will be used, so the N devices can speak N, while G devices can speak G. They don't have to resort to "Esperanto" as with B in order to cooperate.

This means that when the G device is associated but not active, it has no impact at all. When the G devices are active they will consume air time roughly in proportion to the amount of data being transferred. This air time would of course be at the G rate as opposed to the N rate, so in the event that the airwaves are fully saturated (eg by a local file transfer), there would be some reduction in the total Mbps achievable by all devices collectively, but there is no penalty for having the G devices associated.

Confusingly, this appears to conflict with what is stated elsewhere - eg

• "Running a mix of draft 11n and 11b/g clients on the same draft 11n router will reduce speed somewhat for the draft 11n client but reduce the speed of the 11g clients by more than half." at SmallNetBuilder
• "In mixed mode, HT protection requires that 802.11n devices send a legacy preamble, followed by an HT preamble ... These HT protection mechanisms significantly reduce an 802.11n WLAN's throughput, but they are necessary to avoid collisions between older 802.11a/b/g devices and newer 802.11n devices." at TechTarget ANZ

Q: Is having a (draft) 802.11N access point advantageous, even if most or all clients on the network are 802.11G?

A: YES, primarily because 802.11N radios have the benefit of more sophisticated multipath reception capability. They can thereby extend the range and throughput available to G devices to some degree.

Comments

[BlueRaja - Danny Pflughoeft]

"Confusingly, this appears to conflict with what is stated elsewhere" - so how do we know which source to trust? Any links to actual studies to determine that there is/isn't a slow-down? Or were any of those written by someone with qualifications?

[harrymc]

The above article by Slim is at least based on tests, and their analysis seems logical. I believe B routers in the future will become unsupported. For G routers, as Ethernet packets are all the same size, G doing file-transfer at 20 Mbps can slow-down 300 Mbps N by monopolizing the router 15 times longer for same-sized packets, unless the router does time-slicing rather than data-slicing, in effect penalizing G heavily. In the end, network throughput depends on the router's smartness, which must to some degree penalize one or the other (or both), which might explain the conflicts.

[harrymc]

To put it much too simply and using the above numbers: If the router does round-robin, then a G client is equivalent to 15 N clients, reducing the N throughput but conserving G throughput. If the router gives equal time to each, then each N client will do 15 packets for one packet by G client, reducing the G throughput but conserving N throughput. Any router may have its own scheduling algorithm, so giving different results when tested.

[Evelina]

There is some slowdown when 802.11g devices are present - see detailed analysis from a WiFi vendor here: nle.com/literature/…