Receive Window - SSH

This is a discussion on Receive Window - SSH ; Anyone come across the concept of this as it applies to SSH? Apparently the standard default is small and limits speed....

+ Reply to Thread
Results 1 to 7 of 7

Thread: Receive Window

  1. Receive Window

    Anyone come across the concept of this as it applies to SSH? Apparently
    the standard default is small and limits speed.

  2. Re: Receive Window

    >>>>> "PB" == pbarwich writes:

    PB> Anyone come across the concept of this as it applies to SSH?
    PB> Apparently the standard default is small and limits speed.

    http://www.snailbook.com/docs/connection.txt

    Search for "window".

    --
    Richard Silverman
    res@qoxp.net


  3. Re: Receive Window

    Richard,

    OK, I read all that, and thanks for the quick response.
    My question then is how do I optimise a window size in OpenSSH for
    windows when it is contacted from a remote machine by Putty?. With
    window size adjustments in my win2k machine I can comfortably get
    1.4Mbit downloads with around 200mS RTT, but nowhere near that if I run
    the test from a remote machine via SSH.

    Thaks,

    Peter

    Richard E. Silverman wrote:
    >>>>>>"PB" == pbarwich writes:

    >
    >
    > PB> Anyone come across the concept of this as it applies to SSH?
    > PB> Apparently the standard default is small and limits speed.
    >
    > http://www.snailbook.com/docs/connection.txt
    >
    > Search for "window".
    >


  4. Re: Receive Window

    On 2005-11-04, pbarwich wrote:
    > OK, I read all that, and thanks for the quick response.
    > My question then is how do I optimise a window size in OpenSSH for
    > windows when it is contacted from a remote machine by Putty?. With
    > window size adjustments in my win2k machine I can comfortably get
    > 1.4Mbit downloads with around 200mS RTT, but nowhere near that if I run
    > the test from a remote machine via SSH.


    Currently, OpenSSH limits the channel window sizes to around 64kB.
    With a RTT of 200ms, that should be enough to saturate 64k * 8 / .200 =
    2.5 Mbit/s.

    There's a patch at http://www.psc.edu/networking/projects/hpn-ssh/ which
    makes the channel windows dynamic which is worth a try (but be aware
    that in its current form it will also slow some LAN connections too).

    The other thing worth trying is fast cipher/mac pairs. Of the standard
    ones, arcfour and hmac-md5-96 are usually fastest.

    --
    Darren Tucker (dtucker at zip.com.au)
    GPG key 8FF4FA69 / D9A3 86E9 7EEE AF4B B2D4 37C9 C982 80C7 8FF4 FA69
    Good judgement comes with experience. Unfortunately, the experience
    usually comes from bad judgement.

  5. Re: Receive Window

    Darren Tucker wrote:
    > There's a patch at http://www.psc.edu/networking/projects/hpn-ssh/ which
    > makes the channel windows dynamic which is worth a try (but be aware
    > that in its current form it will also slow some LAN connections too).


    And just as a data point, I've used that patch for some WAN connected
    machines. It worked very well (order of magnitude increase) and we have
    not noticed any reduction in performance on the LAN connections. It may
    be slower, but without testing I can't really tell.

    I haven't even bothered to examine if we're normally network or CPU
    bound or what, but since no one's noticed a problem, we just leave it in
    on all the machines now...

    Using -z none instead of the default ciphers gave only a small
    performance increase, so no one pressed to have it enabled. :-)

    --
    Darren Dunham ddunham@taos.com
    Senior Technical Consultant TAOS http://www.taos.com/
    Got some Dr Pepper? San Francisco, CA bay area
    < This line left intentionally blank to confuse you. >

  6. Re: Receive Window

    Darren(s),

    Thank you both for your inputs. I had come across the patch described
    below, but my problem is I'm too much of a newbie to apply it. I got my
    SSH from the windows site, (http://sshwindows.sourceforge.net/), and
    have only used the executables supplied there, with Putty as a client. I
    did come across setup-hpnssh.exe, (available at
    http://www.psc.edu/networking/projec...tup-hpnssh.exe) but
    it's not clear to me that it's giving me a speed advantage, though there
    are some other potential bottlenecks I'm investigating too. It also
    kicks in with a 'bash' shell, which doesn't seem to affect the
    tunelling, but which I'm not familiar with, and don't think I want or
    need in a windows environment.

    If someone can point me at some instructions as to how to patch
    I'll give it a go.

    Many Thanks,

    Peter

    Darren Dunham wrote:
    > Darren Tucker wrote:
    >
    >>There's a patch at http://www.psc.edu/networking/projects/hpn-ssh/ which
    >>makes the channel windows dynamic which is worth a try (but be aware
    >>that in its current form it will also slow some LAN connections too).

    >
    >
    > And just as a data point, I've used that patch for some WAN connected
    > machines. It worked very well (order of magnitude increase) and we have
    > not noticed any reduction in performance on the LAN connections. It may
    > be slower, but without testing I can't really tell.
    >
    > I haven't even bothered to examine if we're normally network or CPU
    > bound or what, but since no one's noticed a problem, we just leave it in
    > on all the machines now...
    >
    > Using -z none instead of the default ciphers gave only a small
    > performance increase, so no one pressed to have it enabled. :-)
    >


  7. Re: Receive Window

    pbarwich wrote:
    > Darren(s),


    > Thank you both for your inputs. I had come across the patch described
    > below, but my problem is I'm too much of a newbie to apply it. I got my
    > SSH from the windows site, (http://sshwindows.sourceforge.net/), and
    > have only used the executables supplied there, with Putty as a client.


    Because sshwindows is a collection of already-compiled tools, I would
    send them a note and ask if they could incorporate the patch into
    something that could be downloaded.

    The patches are source patches and require SSH to be compiled. It's not
    something you can apply to existing binaries.

    Otherwise, I would imagine you'd want to have a full cygwin environment,
    (including gcc and development tools) and compile OpenSSH yourself.
    Once you're happy with that, you can apply the patch and you'll have a
    patched version.

    I've not heard of it before, but it appears that copSSH is a
    distribution similar to sshwindows (I'm not sure what the differences
    are). There is a version of it with the patch integrated.

    http://archive.erdelynet.com/ssh-l/2005-10/3461.html

    I have no idea how well it might work.

    > did come across setup-hpnssh.exe, (available at
    > http://www.psc.edu/networking/projec...tup-hpnssh.exe) but
    > it's not clear to me that it's giving me a speed advantage, though there
    > are some other potential bottlenecks I'm investigating too. It also
    > kicks in with a 'bash' shell, which doesn't seem to affect the
    > tunelling, but which I'm not familiar with, and don't think I want or
    > need in a windows environment.


    Hmm.. I'm not exactly sure what that file is. I wasn't able to find any
    page that describes it (and I didn't download and run it.)

    --
    Darren Dunham ddunham@taos.com
    Senior Technical Consultant TAOS http://www.taos.com/
    Got some Dr Pepper? San Francisco, CA bay area
    < This line left intentionally blank to confuse you. >

+ Reply to Thread