Lost packets

[sinshiva]

well. i've set everything on my lan, every client and interface on every device back to 1500. only left eth0 on the router running at 1456. i've rebooted EVERY device.

everything seems to work, however when i speedtest on my win7 desktop, i consistently get 10.10 down. if i set the client back to 1456, 10.15 and everything is snappier. so, i guess leaving the lan at 1500 works, but i don't seem to benefit from the efficiency of my 1456 mtu over my LLC bridged ADSL connection. for the life of me, i can't explain this behavior. maybe it's something to do with the firmware?

for now, i'm just going to leave the lan side network at 1500, but force the clients back to 1456. i truly hoped these days were over.

PPPoE users, might be useful to set 1492 on the clients

[edit/] well, i've checked every interface on the network about 50 times, all set for 1500 except when i test on a client. i reproduce the results with consistency. pretty weird

 

[PaulnOZ]

Good to hear your setup is working better.

Unfortunately I'm still getting this from ICSI

-----------
The client was able to send fragmented UDP traffic.
The client was unable to receive fragmented UDP traffic. The most likely cause is an error in your network's firewall configuration or NAT.
The maximum packet successfully received was 1464 bytes of payload.

Path MTU (?): Warning –
The path between your network and our system supports an MTU of at least 1500 bytes, and the path between our system and your network has an MTU of 1492 bytes. The path MTU bottleneck that fails to properly report the ICMP "too big" is between [my IP number] and [my IP number]. The path between our system and your network does not appear to report properly when the sender needs to fragment traffic.
------------

Putting the shark on the wire also confirms that packets larger than 1464 marked with the "Don't Fragment" flag set are not making it into my network. Outbound no problem, unflagged large packets no problem in or out, but large inbound flagged packets just never appear.

Can't wait to solve this little mysery.

 

[PaulnOZ]

Well, I'm off to complain to my ISP. It's simply got to be one of their servers.

 

[PaulnOZ]

Further to the saga

My ISP Exetel Australia in their wisdom have decided to follow RFC4638 and set an MTU of 1500 over PPPoE instead of the more normal RFC2516 1492 MTU that most modems and router follow.

My RT-N66U router (through a LLC Bridged n10 modem) doesn't allow an MTU of 1500 over PPPoE and restricts the maximum to 1492 like most CPE stuff does.

While an MTU of 1492 will appear to work fine for most people, the truth is that packets will be dropped and lost however most people probably won't notice or care.

Unfortunately I am loosing data so I was hoping that there was a way to force the router to do 1500 over PPPoE

I believe this might be a kernel limit that was fixed in some newer releases but I'm no Linux expert.

Is it possible to set a 1500 or 1508 MTU over PPPoE on the RT-N66U?

 

[RMerlin]

Not at this time, no. I still need to investigate to see if rp-pppoe does properly support such packets.

 

[PaulnOZ]

Thanks RMerlin,

According to this David said he'd look into it back in Jan 2012. Hopefully something's been done by now. Apparently some others have managed to get it working but I'm sure you already know that.

If there's any way I can help out, just let me know because I have a vested interest in seeing this happen.

And thanks also for the FW RMerlin, really nice work mate.

 

[RMerlin]

Thanks RMerlin,

According to this David said he'd look into it back in Jan 2012. Hopefully something's been done by now. Apparently some others have managed to get it working but I'm sure you already know that.

I vaguely remember seeing a patch being posted online about this, but I haven't had the chance to investigate yet. Just odd that it's not merged in yet if that's really the case, considering 3.11 has been released quite a while ago now. Might be because the patch isn't properly designed, or it introduces other issues.

 

[sinshiva]

well. i've set everything on my lan, every client and interface on every device back to 1500. only left eth0 on the router running at 1456. i've rebooted EVERY device.

everything seems to work, however when i speedtest on my win7 desktop, i consistently get 10.10 down. if i set the client back to 1456, 10.15 and everything is snappier. so, i guess leaving the lan at 1500 works, but i don't seem to benefit from the efficiency of my 1456 mtu over my LLC bridged ADSL connection. for the life of me, i can't explain this behavior. maybe it's something to do with the firmware?

for now, i'm just going to leave the lan side network at 1500, but force the clients back to 1456. i truly hoped these days were over.

PPPoE users, might be useful to set 1492 on the clients

[edit/] well, i've checked every interface on the network about 50 times, all set for 1500 except when i test on a client. i reproduce the results with consistency. pretty weird

finally resolved the weirdness with the following command;

iptables -I FORWARD -p tcp --tcp-flags SYN,RST SYN -j TCPMSS  --clamp-mss-to-pmtu

pretty substantial performance increase when using a lower than 1500 WAN mtu with 1500 mtu clients

[edit1]

using

#!/bin/sh
iptables -I FORWARD -p tcp --tcp-flags SYN,RST SYN -j TCPMSS --clamp-mss-to-pmtu

added to /jffs/scripts/firewall-start
chmod a+rx /jffs/scripts*
[/edit1]

[edit2/]

this is my new favorite iptables trick, heh. i can finally set everything on my lan to 1500 mtu, including all the inside interfaces of the asus and the routers behind it. now, i am finally seeing the same performance as if my client were manually configured with the WAN mtu!!