Router Latency When Using P2P

Discussion in 'Broadband' started by Richard Sacks, Aug 4, 2003.

  1. I'm interested in using a voice-over-IP service, but have noticed that the
    latency on my router increases to about 600 ms when my connection is max-ing
    out under p2p traffic. It shouldn't do: I've only got a 512 connection, but
    the router is specified up to 8 Mbits/s. A 600 ms latency would make VoIP
    call unbearable!

    Is this a specific problem to my router, which has a Connexant chipset, or
    is this common to all low-end routers?

    Thanks,


    Richard
     
    Richard Sacks, Aug 4, 2003
    #1
    1. Advertisements

  2. Perhaps I'm mistaken, but my understanding is that bandwidth and latency are
    different things. Just because the pipe is full doesn't mean that data
    should take longer to get into the pipe.

    Yes, on the uplink, the router needs to make a decision about which packets
    to drop, but why should that decision take a whopping 600 ms?

    Perhaps I shouldn't have used the term "maxing". This effect occurs even
    when the connection is far from full. It's such a problem with KaZaA that I
    don't use that app any more.

    If my router was only specified for 512/256 connections, then that might
    imply that it was designed for less heavy users than myself and its
    performance not being as good as other routers. But it isn't, so I'm still
    confused...


    Richard
     
    Richard Sacks, Aug 4, 2003
    #2
    1. Advertisements

  3. Richard Sacks

    Clueless Guest


    This is a budget router you are talking about, what makes you think it can
    prioritise traffic? One packet might look just the same as the other to it.
    Spend more if you want something capable of deciding what to drop. Look for
    one hat supports Layer 3 switching,
    http://www.synetrix.co.uk/products/voice_data/index.cfm for an overview.
    http://makeashorterlink.com/?K28A45C75 for more Google links.

    S. Althaf
     
    Clueless, Aug 4, 2003
    #3
  4. I don't!
    If I had traffic shaping, then I wouldn't need to post here :)
    This router is so budget that I bet it just dumps every Xth packet. Still,
    that requires numbers to be crunched.
    Likely so, in this case. Only if the router had some intelligence, such as a
    relatively simple weighted round robin scheme, would pay attention to packet
    parameters.


    Regardless, none of this answers my question! Is this a common problem? Is
    600 ms typical?

    Thanks, chaps.


    Richard
     
    Richard Sacks, Aug 4, 2003
    #4
  5. Richard Sacks

    CB Guest

    : Regardless, none of this answers my question! Is this a common problem? Is
    : 600 ms typical?

    I have a conexant based chipset router as well and do not experience
    this. *however* I do not upload at all when using P2P applications and
    am using Routed IP (ie, non NAT).

    I think what you are finding is inherent a problem on TCP/IP - see here:

    http://homepage.ntlworld.com/robin.d.h.walker/cmtips/downup.html
     
    CB, Aug 4, 2003
    #5
  6. Richard Sacks

    CB Guest

    : > It shouldn't do
    :
    : Could you explain as I would have thought, regardless of the specification of
    : the router that you have, as all of your bandwidth is being used for P2P
    : wouldn't that account for the issue rather than a fault with the router as you
    : suggest...?
    :
    It's a fault in TCP/IP:

    http://homepage.ntlworld.com/robin.d.h.walker/cmtips/downup.html
     
    CB, Aug 4, 2003
    #6
  7. I think you're missing the point Richard . Think of the router as a bridge
    with cars going over it - latency being the transit time of a car over the
    bridge, plus the roundabout before it..

    All the cars enter the bridge via the roundabout. Once they clear the
    roundabout, they pass across the bridge.

    If there is a constant flow of traffic across the bridge, such that the
    density of traffic is quite low, then the transit time is constantly low.

    If however you pump so much traffic onto the roundabout (P2P traffic) that
    new traffic joining simply cannot get onto the roundabout, transit times
    increase. The time to cross the bridge is constant, but there is a delay at
    the roundabout.

    You simply cannot exceed the capacity of the bridge, and that¹s what you are
    expecting.

    As to the 600ms - it could be any number at all, or not at all if the
    bridge is full. If I set up an FTP server and access it at full capacity, I
    cannot browse the internet at the same time - requests go out but nothing
    comes back. The lag is therefore much more than 600ms - virtually infinite
    in fact.

    Stop downloading porn long enough to make the call imho :)
     
    Graham in Melton, Aug 4, 2003
    #7
  8. Richard Sacks

    K Guest

    Share some files you stingy bastard.

    K
     
    K, Aug 4, 2003
    #8
  9. Richard Sacks

    CB Guest

    :
    : : > I have a conexant based chipset router as well and do not experience
    : > this. *however* I do not upload at all when using P2P applications
    :
    : Share some files you stingy bastard.

    And have the RIAA and it's co-horts onto me? No thankyou very much.
    Uploading on my connection is hit and miss anyway, my download rate
    slows when I start to upload and the whole connection just goes sloooww.
     
    CB, Aug 4, 2003
    #9
  10. Because it has buffers - that is how it works. I am surprised it is only
    600ms.
    If you send more data than can go out then it is buffered. Buffers are
    often many K in size and 600ms is not bad. This has nothing much to do
    with the speed of the link at all.
    Nothing to do with it. The only impact is that most routers have a fixed
    size buffer and so a faster link is inherently lower max latency.
     
    Rev Adrian Kennard, Aug 4, 2003
    #10
  11. Not a fault - that is exactly how it is designed to work.
     
    Rev Adrian Kennard, Aug 4, 2003
    #11
  12. Richard Sacks

    CB Guest

    : CB wrote:
    : > On Mon, 4 Aug 2003 17:19:52 +0100 and in article <bgm13r$fg4$1
    : > @news8.svr.pol.co.uk>, Morgan said...
    : > : > It shouldn't do
    : > :
    : > : Could you explain as I would have thought, regardless of the specification of
    : > : the router that you have, as all of your bandwidth is being used for P2P
    : > : wouldn't that account for the issue rather than a fault with the router as you
    : > : suggest...?
    : > :
    : > It's a fault in TCP/IP:
    :
    : Not a fault - that is exactly how it is designed to work.

    Ok, correction noted. Can I ask why it works in such a way, or perhaps
    a friendly pointer to reference as to why?

    It seems a bit of a shortsighted implementation :/
     
    CB, Aug 4, 2003
    #12
  13. TCP is trying to maximise the efficient use of the link. This means it
    sends packets at increasing rates until it starts to drop packets and
    then backs off. This makes the absolute most efficient use of all of the
    interviening links and ensures they all have their buffers full.

    IP actually allows for tagging packets "lowest latency", but it is so
    easy to "abuse" that nothing really takes any notice in traffic
    prioritisation, so you are stuck with the way TCP works.

    Suitable equipment - a smart ADSL router or a very smart router in front
    of it like a firebrick, can manage traffic flows and sequence packets to
    do what you want at your end and as that is likely to be the bottleneck
    for uplink it is useful.
     
    Rev Adrian Kennard, Aug 4, 2003
    #13
  14. Makes for a nice diagram on the whiteboard on the IP courses we run as
    well <-:
     
    Rev Adrian Kennard, Aug 4, 2003
    #14
  15. Richard Sacks

    CB Guest

    : CB wrote:
    : > On Mon, 04 Aug 2003 19:57:54 +0100 and in article <bgmabi$q2d3b$5@ID-
    : >...
    : > : Not a fault - that is exactly how it is designed to work.
    : >
    : > Ok, correction noted. Can I ask why it works in such a way, or perhaps
    : > a friendly pointer to reference as to why?
    : >
    : > It seems a bit of a shortsighted implementation :/
    :
    : TCP is trying to maximise the efficient use of the link. This means it
    : sends packets at increasing rates until it starts to drop packets and
    : then backs off. This makes the absolute most efficient use of all of the
    : interviening links and ensures they all have their buffers full.
    [snip excellent information]

    As always Adrian you are a wealth of information. Appreciate your
    comments.

    /me goes to read up more on this
     
    CB, Aug 4, 2003
    #15
  16. Richard Sacks

    CB Guest

    :
    : > And have the RIAA and it's co-horts onto me? No thankyou very much.
    : > Uploading on my connection is hit and miss anyway, my download rate
    : > slows when I start to upload and the whole connection just goes sloooww.
    :
    : Are the RIAA taking legal action outside the U.S. then?

    Not that I am aware of.

    : I thought they only
    : have a remit for the U.S....

    They do, however, RIAA has brances of its interests all of the Western
    World. I dont really use P2P but the times I do i'm fucked if i'm going
    to *remotely* risk their attention. Sorry dudes, but self preservation
    comes first!
     
    CB, Aug 5, 2003
    #16
  17. That is indeed a good analogy. I understand the reasoning now.

    Any idea how big typical buffers are, in terms of the maximum latency they
    can introduce?
     
    Richard Sacks, Aug 5, 2003
    #17
  18. Are the RIAA taking legal action outside the U.S. then? I thought they only
    have a remit for the U.S....
     
    Richard Sacks, Aug 5, 2003
    #18
  19. Switching, or routing?
    Regardless, do you mean (de)prioritising by port number?

    But don't the latest versions of p2p apps use random port numbers to avoid
    detection/ISP port throttling?
     
    Richard Sacks, Aug 5, 2003
    #19
  20. They're only regional licensing representatives though.

    I take your point tho, better safe than sorry.
     
    Richard Sacks, Aug 5, 2003
    #20
    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.