Wireless MTU modification

systemcrash · August 6, 2020, 12:26pm

I read this thread and tried recommendations here (but this may apply to older LEDE/OWRT builds).

How does one modify Wireless MTU? There is currently no GUI option ( I don't mind making a PR to add it ) - although trying in /etc/config/wireless yields no change if I supply
option mtu 2000 under config wifi-iface 'wifinet0' or config wifi-device 'radio0' - any tips?

It results in
31: wlan1: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 qdisc noqueue state DOWN group default qlen 1000 link/ether xx:xx:xx:df:39:72 brd ff:ff:ff:ff:ff:ff

I also tried the Fragmentation Threshold advanced wifi option. Nothing there either. Some hardware is just hard-coded/cannot do it?

TP-Link Archer C7 v2

pwned · August 6, 2020, 12:54pm

That has to be supported by hardware/driver on both sides of the connection with the same size configured. Otherwise it will be fragmented. It usually has to be enabled explicit. Also I doubt you will see any effect nowdays. CPUs even on routers are powerfull enough. Beside that there is no standard for jumbo frames. So each device/driver is supporting sizes what it want.

systemcrash · August 11, 2020, 5:54pm

You're right: there is no standard for jumbo frames - in Ethernet context though newer Ethernet devices support 9K, and some 65K.

SO Answer on MTU sizes

Tech article on max throughput

The article mentions MPDU aggregation increases the maximum frame size from 2304 octets to 65535. It seems this is supported in more modern 802.11e-n-ac standards.

It seems that these larger sizes are very much defined - it would be nice to know whether we are utilizing them, and/or able to.

802.11 standards define MSDU size in DMG networks as 7920 bytes.

The future is coming, >= 40GHz devices will open this up, but I suppose those devices will drive this development in the GUI

See

github.com

torvalds/linux/blob/00e4db51259a5f936fec1424b884f029479d3981/include/linux/ieee80211.h#L223


#define IEEE80211_MAX_MESH_PEERINGS	63
/* Maximum size for the MA-UNITDATA primitive, 802.11 standard section
   6.2.1.1.2.

   802.11e clarifies the figure in section 7.1.2. The frame body is
   up to 2304 octets long (maximum MSDU size) plus any crypt overhead. */
#define IEEE80211_MAX_DATA_LEN		2304
/* 802.11ad extends maximum MSDU size for DMG (freq > 40Ghz) networks
 * to 7920 bytes, see 8.2.3 General frame format
 */
#define IEEE80211_MAX_DATA_LEN_DMG	7920
/* 30 byte 4 addr hdr, 2 byte QoS, 2304 byte MSDU, 12 byte crypt, 4 byte FCS */
#define IEEE80211_MAX_FRAME_LEN		2352

/* Maximal size of an A-MSDU that can be transported in a HT BA session */
#define IEEE80211_MAX_MPDU_LEN_HT_BA		4095

/* Maximal size of an A-MSDU */
#define IEEE80211_MAX_MPDU_LEN_HT_3839		3839
#define IEEE80211_MAX_MPDU_LEN_HT_7935		7935

pwned · August 11, 2020, 7:51pm

Well if there is a standard defined for MSDU size then I'm in hope that industry will make use of it. But I think this will be a premium feature. So they would optimize this values to their own technology/equipment first targeting companies with wider areas in competition with 5G. IMO It will take years hitting the consumer/SOHO market.

One problem you can run into with Wi-Fi is frame payloads can be larger than the standard ethernet payload.

This is IMO the most important point. Like I said already above (fragmentation).
At some point traffic is hitting WAN interface also.
IMO is the usecase for packet size optimisation limited to the network you are running your services (beside the internet using PathMTU). So from the point of an enduser/private user it does not really much for us. Expect you are a heavy LAN user. I might be wrong so don't take that what I've written as in stone written.