Difference between MTU (maximum transmission unit for a TCP/IP packet) versus the fragmentation threshold setting when using 802.11 networks?

They're mutually exclusive from each other. Fragment threshold is access-point centric (802.11), MTU is IP header based..... Difference between layer2 and layer3 in a nutshell.

Yes, fragment threshold can be above MTU, and MTU can be above fragment threshold.

A bit more detail below if you wish to read.


Fragmentation threshold is between your access point and you. If your MTU is lower than your fragment threshold then YOU WILL NEVER see 802.11 fragments.

If your MTU is higher than your frag threshold, AND the payload is large enough then your access-point will force a fragment and the frame will be split up between host and AP to be reassembled upon delivery.

You generally want your MTU to be whatever the maximum MTU is of your AP/LAN/Internet connection. This is generally 1500 bytes, however in some cases where PPPOE/ATMOE/VPLSOE or other carrier-to-carrier encapsulation is used.

Why would you want fragmentation?

The only reason you would be interested in fragmentation is if your wireless network is in a noisy area (alot of other AP's), you see ALOT of loss during an action such as a file transfer, and/or you have a large amount of workstations in your WIFI network.

The reason for this is fairly straightforward:
If your MTU is 1500 bytes and you're receiving data over your wireless network and there's a wireless collision caused by another transmitting station or otherwise, the entire frame is lost and the source station will have to retransmit it (i.e. no ACK received, retransmission) This greatly slows throughput if it happens often.

If your MTU is 1500 bytes, and you fragment threshold is 576 bytes [mere examples], then your AP will force fragments between itself and the wireless device. The payload from the AP is split into multiple 576 byte chunks (3 wireless frames), and are then subsequently reassmbled.

The reason this CAN speed things up, is in a wireless network where RTS/CTS aren't used [this is really the recommended approach], a wireless workstation needs 34us to sense a channel, 16us RTS, data transmission of roughly 600us for full payload, 16us for ack, 34us for DIFS, and roughly another 9us for another slot time.

With a fragmented payload, the time-in flight for a full frame can be reduced, as well as the DIFS time. (i.e. -400us for payload, -34us for DIFS). This also allows other devices to essentially interleave their transmissions within available slots.