[SOLVED] Unable to automatically set the MTU

[POSIX_ME_HARDER]

I'm experiencing network issues between two computers connected by ethernet LAN (and yet no internet issues, it's just when they want to talk to each other).
Those seem to disappear when changing the MTU settings to a lower value, 1400 instead of 1500, but I'm having trouble getting this done automatically.
Indeed, every time I reboot the computer it goes back to 1500 (considering this greatly impedes SSH, it gets quite annoying).
Following instructions from a thread concerning a similar problem, I tried:

Setting the MTU from /etc/dhcpcd.conf:
I Added the following line to the top of /etc/dhcpcd.conf:

[kurly]

Smells like bug 555974... does using >=net-misc/dhcpcd-6.9.2 fix the problem?

My apologies if this winds up being a dead end or non-solution.

[POSIX_ME_HARDER]

Thank you for the suggestion. Sadly, using dhcpcd 6.9.2 did not solve the issue.
In the bug report you linked, dhcpcd seems to be able to change the MTU, but restores it to its previous value right away.
In my case, it seems dhcpcd can't set the MTU because the interface is already up.
Is there perhaps a way to tell dhcpcd that it's OK to turn eth0 off to edit the MTU?

My best solution so far is adding a script in /etc/local.d/, but the fact that it doesn't work the way it's supposed to bothers me a bit.

[krinn]

Sorry, the best thing to do is looking why your two computers have trouble at 1500 MTU.

MTU are standards values, for eth2 it's 1500 (no more, and no less, exactly that value), pppoe use 1492 (because 1942 for eth2 datas + 8 for ppp headers = 1500 the standard MTU of eth2)

Getting out the of the standard you are not solving your issue like you think, you are adding trouble to your trouble, and you lesser to get any help from anyone and you'll be alone to handle this ; even i know the symptoms of someone silly doing that, i have no clue howto myself handle that, except getting back to the standard ; but if you know someone that understand the MTU mechanic deeply, that must include, how any devices (including the device you have no control over) receiving the packets can be alter to handle it with your non standard way, better make friend with it asap!.

Here's symptoms example, and you can see this user did like you lowering MTU to prevent the MTU trouble his router was doing.
Back to 1500 and really dig at who was causing problem, it has fix it. https://forums.gentoo.org/viewtopic-t-935256-highlight-mtu.html

Another one, if you like to see where your "i'm altering the MTU because i can" way will drive you : https://forums.gentoo.org/viewtopic-t-566869-highlight-mtu.html

[NeddySeagoon]

POSIX_ME_HARDER,

The MTU is unlikely to be your problem. In any case, if its sub optimum, things mostly still work but you get extra packets sent over the link as they need to be broken up. This increases the overhead ... until you get a packet with the do not fragment flag set, so it cannot be sent at all.

First, use ping with various packet sizes and the do not fragment flag set to determine the MTU over the link.
If its 1500 (including headers) the MTU is not your issue.

Its rare to need to change the MTU manually. A few old brain dead domestic routers, don't do PPPoE properly and for them, forcing a MTU of 1492 helps.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[POSIX_ME_HARDER]

Sure, let's try something else . I focused on the MTU because it seemed to solve the problem but it could indeed be caused by something else.
I think I'll have to give more information about the connection issues then:
My local network has three computers. Two of which (200 and 202) are connected through ethernet and one (201) using WiFi.
200 and 202 have a hard time talking to each other (trying to connect through SSH will hang at "debug1: expecting SSH2_MSG_KEX_ECDH_REPLY" and distcc gets stuck).
They have no issue communicating with 201 however.
Just in case this could be linked (no pun intended), the modem router connecting them gets its Internet from PPPoE (with a 1492 MTU setting), to get to the Internet the modem-router goes through a second router that will handle packet tagging in and out of the VLAN used by my ISP (it's not optimal, but the modem-router can't do it by itself).

Here are the results of my ping tests:

• 200 <-> 202: If I go higher than 1459 (meaning 1487 total size), there are no replies.
  
• 200 <-> 201: I can go up to the maximum size (1500 total, as expected).
  
• 201 <-> 202: Same here.
  
• 200 <-> Internet: It tells me the maximum MTU is 1492 (as expected? considering how the router is configured...).
  

The command I used is:

[NeddySeagoon]

POSIX_ME_HARDER,

On the face of it

[POSIX_ME_HARDER]

Here are the current routing tables:

• 200:
  

  Code:

  Destination     Gateway         Genmask         Flags Metric Ref    Use Iface 0.0.0.0         192.168.1.1     0.0.0.0         UG    202    0        0 eth0 127.0.0.0       0.0.0.0         255.0.0.0       U     0      0        0 lo 192.168.1.0     0.0.0.0         255.255.255.0   U     202    0        0 eth0

  
  
• 201:
  

  Code:

  Destination     Gateway         Genmask         Flags Metric Ref    Use Iface 0.0.0.0         192.168.1.1     0.0.0.0         UG    303    0        0 wlan0 127.0.0.0       0.0.0.0         255.0.0.0       U     0      0        0 lo 192.168.1.0     0.0.0.0         255.255.255.0   U     303    0        0 wlan0

  
  
• 202:
  

  Code:

  Destination     Gateway         Genmask         Flags Metric Ref    Use Iface 0.0.0.0         192.168.1.1     0.0.0.0         UG    2      0        0 eth0 127.0.0.0       0.0.0.0         255.0.0.0       U     0      0        0 lo 192.168.1.0     0.0.0.0         255.255.255.0   U     2      0        0 eth0

  
  

I'll try connecting 202 to 200 directly and post the results ASAP (it may take a while).

[grosgood]

Curious that the observed problem coincides with the transition from Arch --> Gentoo on 202. Could this have involved physical movement of the machine? A general cleaning of hardware? Vacuuming out dust bunnies? Pulling and reinserting of network cables? A dirty RJ45 plug can host all kinds of network gremlins that don't wake up until the cable is physically disturbed and dust gets in between those little wires.

When you do Neddy Seagoon's test, don't be surprised if the problem disappears. Or maybe put a fresh set of shiny new cables in your present setup and rerun your tests. I've fixed network problems doing nothing more than that.

Good luck!
_________________
Garry

[NeddySeagoon]

POSIX_ME_HARDER,

Your wired ethernet network topology is a 'V' shape, with 200 and 202 on the top of the arms and the router at 192.168.1.1 at the bottom of the 'V'
Its a bit scary that 201 is using WiFi on the same subnet as wired, both from a security point of view and and a routing pount of view. It can be made to work though.

There is nothing there to contribute to your problem.


grosgood,

Its called 'wiping the contacts'. If this is the issue, ifconfig should show errors.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[POSIX_ME_HARDER]

Sorry for the delay, I had never linked two computers directly.

Yup, as you both expected, the problem does not appear when connecting 200 and 202 to each other directly, the MTU is indeed 1500.

I'm surprised the local network topology is considered weird (what I consider weird about my network is the router whose sole role is to tag packets), it's what most ISPs boxes do, right? encrypted WiFi + ethernet.

And yes, I cleaned up 202 just before installing Gentoo. I'll do it again, just in case there's dust in the RJ45 port (it's indeed possible, as the board was quite dusty).

Edit:
Yup, cleaning up 202 again resolved the problem. Thank you very much to you guys for your time, glad to see this forum actively helps its members. I'll now procede to laughing out loud for a few more minutes about the hours spent trying to fix what was actually just some dust.

[NeddySeagoon]

POSIX_ME_HARDER,

That leaves your cables and your router at 192.168.1.1.

I'm supposed to use Big Thiefs Network Termination Equipment as I'm the UK.
Thats a transparent VDSL to PPoE bridge. I don't need to set the VLAN as its in BTs NTE.

However, if use my own VDSL modem, I need to set the VLAN or I don't get any data.
There is a lot to be said for using my own VDSL modem but as my data rate is at the system provided maximum, I don't bother. Not often anyway.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[POSIX_ME_HARDER]

I should have done a better test than a simple SSH connection: turns out the reason it works is because 202 rebooted after I cleaned it up and reapplied the local script (I forgot about) to set a 1400 MTU. Now I'm confused, why did it allow 1500 MTU pings earlier (when connecting 200 and 202 directly), despite the fact that the script was still there, I have no idea.

Still, I'll redo the test once more, just to be sure.

I've connected 201 to the network using ethernet instead. Surprise: it suffers the same issue as the others, so clearly the computers on my local network can't communicate using both ethernet and MTU 1500.

Now, let's get to the modem... (although I don't understand why a modem problem would not appear when I was using Archlinux ARM).
It's a "TP-Link ADSL2+ Gigabit Double Bande AC1750", with the firmware "0.9.1 0.12 v002d.0 Build 150514 Rel.33266n", the ISP is "Orange" (France).
I'm connected using EWAN (PPPoE), with a MTU of 1492.
I have no idea what we're looking for.

[NeddySeagoon]

POSIX_ME_HARDER,

The routing table from the router would be good

When I first got ADSL, my bandwidth over my network suddenly dropped.
All my local traffic was getting sent over my ADSL link to my ISP and bacK.
Oops. It took me a few days to discover that :)

Even then, a MTU of 1492 should work. Is your router one of these?
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[POSIX_ME_HARDER]

Yes, it is. I have an optic fiber though (I doubt it changes anything concerning this issue).
I tried using Windows on 200 to ping 202, same results. I don't understand how it could work with Arch before. yet I know it did.
I'll create a thread on TP link's forums as well.

Oh, and since I forgot to do the tests before:

200 <-> modem: 1500 MTU max.
202 <-> modem: 1500 MTU max.

[NeddySeagoon]

POSIX_ME_HARDER,

Try tracert and see where the traffic goes.
It should go to the router then to the other PC.

Fibre and only ADSL2+ Ewww.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[POSIX_ME_HARDER]

[NeddySeagoon]

POSIX_ME_HARDER,

I used to get 4Mbit/sec over ADSL2+ without fibre.
I now get 80Mbit/sec with fiber to the cabinet. The last 40m uses the phone line for VDSL.
All the hardware is 100Mbit/sec capable but BT won't let us use it even for money.
While your router is ADSL2+ capable, you are not using the ADSL2+ port for the internet connection. That was confusion on my part.
Sorry about that.

The VLAN may increase your latency but it will not affect youl data rate.

I expected to see your router listed there too.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[krinn]

[NeddySeagoon]

krinn,

A router has two sides an upstream side (WAN) and a downstream side (LAN).
If the WAN port is a PPPoE endpoint then 1492 is correct on the WAN.
As you say, the LAN side should be 1500.

Route MTU discovery will use 1492 going to the WAN and 1500 to the LAN.

The router may have both 1492 and 1500 MTUs configured on the WAN/LAN.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[POSIX_ME_HARDER]

I tried changing the setting for the PPPoE, 1492 is the maximum value (the interface won't let me go higher), setting it to a lower value does not affect the MTU between PCs of the local network (it does however lower it when testing towards the Internet, as expected).

The router between the TP-Link and the ONT is a "Netgear GS108E v3", it's configured to tag packets comming from the TP-Link and untag those coming from the ONT.

[krinn]

And jumbo frames support is disable?

[gordonb3]

May I assume that you have the ethernet connected machines hooked up to the TP-Link router? How about using the netgear switch instead? You probably only use 3 of its 8 ports for NT (tagged), internet (untagged) and television (untagged). Simply create a new VLAN ID (e.g. 100) and let the remaining ports be a member of this VLAN ID only. These will now act as a 5(?) port switch that you can use for your LAN.

[UberLord]

[POSIX_ME_HARDER]

I'm unsure how to check for the jumbo frames.
202 can only go up to 100MBit/s, jumbo frames can only be used on gigabit ethernet, right?
I tested on 200 by setting the MTU to 2000 and pinging the modem-router. I can send packets of (total) size up to 1505 bytes, so I'm guessing that means jumbo frames aren't enabled.