0Optimising WAN MTU for BT Infinity

Since switch­ing from my BT SmartHub to a pfsense router I’ve had a few minor teeth­ing prob­lems. One of which has been the squid proxy ser­vice caus­ing a prob­lem where I can­’t access the BBC web­site with an error. Whilst I did even­tu­ally fix this through a slightly strange meth­od, I encountered a pos­sible source of the prob­lem which was worth fix­ing on it’s own

When I turned squid off I was imme­di­ately able to access the BBC web­site, but with it turned on I con­sist­ent received an error from squid that there had been a “Zero Sized Reply”. It seemed squid was unable to get any reply from the BBC. One pos­sible cause of this could be frag­men­ted pack­ets being dropped, and the cause of frag­ment­a­tion is nor­mally MTU size.

The default MTU for Eth­er­net is 1500 and most mod­ern oper­at­ing sys­tems man­age things on the LAN auto­mat­ic­ally. It is pos­sible to use sizes up to 9000 to increase through­put (slightly) for the LAN. How­ever, this was of less interest to me than mak­ing sure pack­ets wer­en’t get­ting frag­men­ted mul­tiple times, or being frag­men­ted after leav­ing the router.

With the pfsense WAN MTU set to default (1500) I tried some pinging from win­dows to see what the max­im­um size was without frag­ment­a­tion. The com­mand to use is ping google.com -f -l 1500, chan­ging the last num­ber to find the optim­um. My BT Infin­ity con­nec­tion uses PPPoE which typ­ic­ally has a max­im­um value of 1492 (it may in fact really use PPPoA which my bridged modem can present as PPPoE to pfsense, dif­fer­ent sources say dif­fer­ent things about this), and when using ping it is neces­sary to take into account some over­head which has a value of 28, so I was expect­ing a ping MTU of 1464 to be the largest I could use. To my sur­prise this still res­ul­ted in frag­ment­a­tion and I even­tu­ally found the max­im­um I could use was 1456.

I used this value (with 28 added back on) to set a MTU value of 1484 in the WAN inter­face set­tings of pfsense. Now the only frag­ment­a­tion should be done by the router to LAN pack­ets, with no risk of any later frag­ment­a­tion from a router some­where between me and any par­tic­u­lar web­site. After clear­ing my browser cache and the squid cache I was then able to load the BBC web­site suc­cess­fully. I sus­pect the issue was caused by a router black­hol­ing issue some­where between my ISP and the BBC serv­ers which was drop­ping the frag­men­ted pack­ets. A nice descrip­tion of the phe­nomen­on can be found on RCL’s Blog