How common are MTU problems?

Discussion in 'Broadband' started by Peter, Aug 9, 2015.

  1. Peter

    Peter Guest

    I have seen various cases of this.

    One of them, my company website was not accessible via Vodafone UK 3G.
    The site is hosted on an ADSL-connected FreeBSD server and the ISP is
    Andrews and Arnold.

    The same website sitting on an identical backup server, where the ISP
    is ZEN, worked fine.

    That one was solved by dropping the MTU on the router from 1492 to
    1442. That exact max value was established by running a ping from a
    laptop on a Vodafone 3G WAN connection.

    Later, after finding many extremely slow websites at work (A&A), we
    dropped the MTU to 1400 and then to 1300 (this value being configured
    in the Draytel 2955 router) and it made an instant and dramatic
    improvement. But the said websites did always work, eventually. One of
    them was the Post Office one (where you type in the daily postage
    stuff).

    I am now looking at another one which according to the host (a fast
    server in Germany and run by a friend of mine) has an MTU of 1500. It
    is not accessible over a Thuraya XT GMPRS satellite phone connection
    (the client is a Lenovo win8 tablet, connected via USB). The owner
    dropped the MTU to 1400 which didn't help but he was of the view that
    this is wrong and inefficient and should not be necessary so no more
    testing was done. But another version of the server code, running on
    the above mentioned 1300-MTU server at my office, works perfectly from
    Thuraya. I don't know if there is any relevant config on that win8
    tablet - never seen MTU anywhere in Windows and I am sure the
    *outgoing* packets in this case are very small.

    I have been doing comms (embedded systems) since the 1970s and do know
    that smaller packets get through more likely on a noisy line, but
    these are very specific values.

    Clearly the MTU negotiation is failing somewhere along the line...

    The internet is packed with reports of such and such website being
    inaccessible, etc...
     
    Peter, Aug 9, 2015
    #1
    1. Advertisements

  2. Peter

    Graham J Guest

    Peter wrote:

    [snip]
    He is of course right.

    Which does not mean that everybody else who configures 3G, satellite, or
    whatever is right. In fact I suspect that many such systems are not
    properly configured.

    So take the pragmatic view, that if you have to reduce the MTU to get
    certain services to function correctly, then do so.

    I can see that your friend in Germany does not want to compromise the
    performance of his server. Can he provide a second server running on a
    separate host, with an MTU of less than 1400?

    I don't know how you would direct traffic arriving from the satellite
    phone to the alternate site - perhaps others here can suggest something.
     
    Graham J, Aug 9, 2015
    #2
    1. Advertisements

  3. Peter

    Peter Guest

    One obvious way would be for the server in our office to act as a
    gateway for the non-working one, but I am not really too concerned
    because I have alternatives.

    It's just weird that this is so common.
     
    Peter, Aug 9, 2015
    #3
  4. Peter

    Stephen Guest

    Sort of....
    TCP overhead on 1500 byte packets is 20 bytes of TCP, 20 bytes of IP,
    18 bytes of Ethernet, so 3.8% and change

    TCP overhead on 1300 bytes 4.4%+

    (substitute your own numbers for your favorite protocol, and VLAN tags
    etc)
    so around 0.6% added overhead

    - if that makes a big difference to a server it is already close to
    crash and burn.......
    Stephen Hope
    Replace xyz with ntl to reply
     
    Stephen, Aug 9, 2015
    #4
  5. Peter

    Andy Furniss Guest

    Different kit/settings in between the server and the WAN? I am suprised
    A&A would mess up where zen didn't. It's possible I suppose that their
    suppliers do mess up - I would have thought A&A would like to know this.
    MTU issues can be perplexing - I am no expert at all and have seen posts
    I can't explain, just some thoughts -

    "Setting MTU on the router" = ambiguous -

    Which interface.

    If you set low MTU on LAN interface it's a good way to break/test for
    all those sites that don't manage to get ICMP frag needed.

    I'll assume you don't do that though :)

    What does the router do - if on wan, does it take a hint and mss clamp
    to this (likely).
    Does it do it properly! You would hope so but I've historically seen
    code that "just sets it" rather than clamping.
    Does it take your wan setting and ask ppp to ask for mru that size -
    this can hurt, though in theory if it mss clamps to match it should work.

    mss clamping affects the size of incoming TCP packets and it's incoming
    size that often causes issues, rather than the size of outgoing.
    Devices tend to look at the mtu of their local interface and use this to
    say in the TCP connection what size (incoming) segment they can take, so
    you don't just affect outgoing size by changing MTU. On routers it's
    more complicated as they are passing traffic and may or may not adjust
    (mss clamp) the tcp maximum segment size.

    There are server settings that after failing to send full size packets,
    after some timeout start sending smaller ones - this could account for
    slow sites.
    There isn't really such a thing as MTU negotiation - it relies on ICMP
    at "run time" and if as some do, servers don't get these due to blocking
    then you have an issue with that server. Who blocks may be out of your
    control, so setting low is a workaround.

    It may also be the case that messing around with the MTU on routers (or
    the way they do things anyway) can set you up to hit issues. If for
    whatever reason "you" can't take a 1500 IP packet, there is a chance you
    will need to send an ICMP frag needed out to a server - which may never
    get it.

    Using sat/3G no doubt complicates things - if it is they that dictate a
    low incoming MTU, then you (your kit) needs to work around it. As I
    don't use such things I just don't know how bad or variable
    things/networks are.

    I've been lucky in that I've always been able to just use 1500 on
    ADSL/FTTC and had ISPs where it just works.
     
    Andy Furniss, Aug 10, 2015
    #5
  6. Peter

    Peter Guest

    Exactly my take on it!
     
    Peter, Aug 10, 2015
    #6
  7. Peter

    Peter Guest

    I could not find anything different.
    I told them - they didn't want to discuss it much at the time. Some of
    the people there are very helpful and some are the exact opposite,
    IME.
    The ADSL one.
    That I don't know.

    We assumed that setting the MTU in the router to say 1442 limits
    outgoing packets to 1442, and attempts to negotiate incoming ones down
    to 1442 also.
    Yes, very likely.
    Sure. I guess somebody may be blocking pings (which is fairly common)
    and accidentally blocking everything all ICMP.
    The additional problem with satcomms (especially Thuraya -
    geostationary) is the long latency. That breaks a lot of interactive
    sites for example.
    It does seem to work OK on ADSL, in so far as no site is IME totally
    inaccessible. But some are just slow.
     
    Peter, Aug 10, 2015
    #7
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.