SLOW sockets under Redhat 9.0 - Questions

This is a discussion on SLOW sockets under Redhat 9.0 - Questions ; I have an application that uses a simple tcp/ip socket to communicate between 2 processes. This has worked fine until we upgraded to RH9.0. Now the socket connection is incredibly slow! What used to take around 5ms now takes 20-90ms! ...

+ Reply to Thread
Results 1 to 9 of 9

Thread: SLOW sockets under Redhat 9.0

  1. SLOW sockets under Redhat 9.0

    I have an application that uses a simple tcp/ip socket to communicate
    between 2 processes. This has worked fine until we upgraded to RH9.0.
    Now the socket connection is incredibly slow! What used to take around
    5ms now takes 20-90ms!

    Anybody have any ideas?

    I've checked all of my calls to make sure I'm not using anything that's
    deprecated (I'm not) and I've tried changing the calls even though this
    setup's been working for over 3 years since we first ported our app from
    the SGI to RH6.1.

    Thanks for any help!!

    David J. Bakeman
    dbakeman@comcast.net


  2. Re: SLOW sockets under Redhat 9.0

    David J. Bakeman wrote:
    > I have an application that uses a simple tcp/ip socket to communicate
    > between 2 processes. This has worked fine until we upgraded to RH9.0.
    > Now the socket connection is incredibly slow! What used to take around
    > 5ms now takes 20-90ms!
    >
    > Anybody have any ideas?
    >
    > I've checked all of my calls to make sure I'm not using anything that's
    > deprecated (I'm not) and I've tried changing the calls even though this
    > setup's been working for over 3 years since we first ported our app from
    > the SGI to RH6.1.
    >
    > Thanks for any help!!
    >
    > David J. Bakeman
    > dbakeman@comcast.net
    >


    What network card are you using? Are you using the exact same drivers
    used in RH8 or was an upgrade necessary? I know NVIDIA network driver
    261 is terrible, and that is what most people that go to RH9 will change to.

    --
    Respectfully,


    CL Gilbert

    "Verily, verily, I say unto you, He that entereth not by the door() into
    the sheepfold{}, but climbeth up some other *way, the same is a thief
    and a robber." John 10:1

    GnuPG Key Fingerprint:
    82A6 8893 C2A1 F64E A9AD 19AE 55B2 4CD7 80D2 0A2D

    For a free Java interface to Freechess.org see
    http://www.rigidsoftware.com/Chess/chess.html


  3. Re: SLOW sockets under Redhat 9.0

    On Thu, 01 Jan 2004 02:33:40 +0000, David J. Bakeman wrote:

    > I have an application that uses a simple tcp/ip socket to communicate
    > between 2 processes. This has worked fine until we upgraded to RH9.0.
    > Now the socket connection is incredibly slow! What used to take around
    > 5ms now takes 20-90ms!
    >
    > Anybody have any ideas?


    There is not enough information to tell what is happening exactly.
    My guess would be that it's related to O(1) with NPTL somehow,
    however it's hard to say without knowing details. It can be a driver,
    as other person mentioned.

    If you could share the source and your testing procedure, it would
    be useful. If you have a good test case, it can be escalated all
    the way to Ingo, no problem.

    -- Pete


  4. Re: SLOW sockets under Redhat 9.0

    Pete Zaitcev (OTID1) wrote:
    > On Thu, 01 Jan 2004 02:33:40 +0000, David J. Bakeman wrote:
    >
    >
    >>I have an application that uses a simple tcp/ip socket to communicate
    >>between 2 processes. This has worked fine until we upgraded to RH9.0.
    >>Now the socket connection is incredibly slow! What used to take around
    >>5ms now takes 20-90ms!
    >>
    >>Anybody have any ideas?

    >
    >
    > There is not enough information to tell what is happening exactly.
    > My guess would be that it's related to O(1) with NPTL somehow,
    > however it's hard to say without knowing details. It can be a driver,
    > as other person mentioned.


    The 2 processes are almost always on the same machine. The sever proc
    creates a AF_INET SOCK_STREAM followed by listen then accept. The
    client proc does a connect after creating it's socket. Both procs then
    set their sockets to TCP_NODELAY.

    >
    > If you could share the source and your testing procedure, it would
    > be useful. If you have a good test case, it can be escalated all
    > the way to Ingo, no problem.


    Unfortunately this is for work so I can't give out the source but I'll
    see if I can create a test programn that has the same problem.

    >
    > -- Pete
    >



  5. Re: SLOW sockets under Redhat 9.0



    David J. Bakeman wrote:
    > Pete Zaitcev (OTID1) wrote:
    >
    >> On Thu, 01 Jan 2004 02:33:40 +0000, David J. Bakeman wrote:
    >>
    >>
    >>> I have an application that uses a simple tcp/ip socket to communicate
    >>> between 2 processes. This has worked fine until we upgraded to
    >>> RH9.0. Now the socket connection is incredibly slow! What used to
    >>> take around 5ms now takes 20-90ms!
    >>>
    >>> Anybody have any ideas?

    >>
    >>
    >>
    >> There is not enough information to tell what is happening exactly.
    >> My guess would be that it's related to O(1) with NPTL somehow,
    >> however it's hard to say without knowing details. It can be a driver,
    >> as other person mentioned.

    >
    >
    > The 2 processes are almost always on the same machine. The sever proc
    > creates a AF_INET SOCK_STREAM followed by listen then accept. The
    > client proc does a connect after creating it's socket. Both procs then
    > set their sockets to TCP_NODELAY.
    >
    >>
    >> If you could share the source and your testing procedure, it would
    >> be useful. If you have a good test case, it can be escalated all
    >> the way to Ingo, no problem.

    >
    >
    > Unfortunately this is for work so I can't give out the source but I'll
    > see if I can create a test programn that has the same problem.
    >
    >>
    >> -- Pete
    >>

    >


    I did ask if you were running an nvidia network card? are you?

    clg


  6. Re: SLOW sockets under Redhat 9.0

    Carl wrote:
    >
    >
    > David J. Bakeman wrote:
    >
    >> Pete Zaitcev (OTID1) wrote:
    >>
    >>> On Thu, 01 Jan 2004 02:33:40 +0000, David J. Bakeman wrote:
    >>>
    >>>
    >>>> I have an application that uses a simple tcp/ip socket to
    >>>> communicate between 2 processes. This has worked fine until we
    >>>> upgraded to RH9.0. Now the socket connection is incredibly slow!
    >>>> What used to take around 5ms now takes 20-90ms!
    >>>>
    >>>> Anybody have any ideas?
    >>>
    >>>
    >>>
    >>>
    >>> There is not enough information to tell what is happening exactly.
    >>> My guess would be that it's related to O(1) with NPTL somehow,
    >>> however it's hard to say without knowing details. It can be a driver,
    >>> as other person mentioned.

    >>
    >>
    >>
    >> The 2 processes are almost always on the same machine. The sever proc
    >> creates a AF_INET SOCK_STREAM followed by listen then accept. The
    >> client proc does a connect after creating it's socket. Both procs
    >> then set their sockets to TCP_NODELAY.
    >>
    >>>
    >>> If you could share the source and your testing procedure, it would
    >>> be useful. If you have a good test case, it can be escalated all
    >>> the way to Ingo, no problem.

    >>
    >>
    >>
    >> Unfortunately this is for work so I can't give out the source but I'll
    >> see if I can create a test programn that has the same problem.
    >>
    >>>
    >>> -- Pete
    >>>

    >>

    >
    > I did ask if you were running an nvidia network card? are you?

    No it's not the nvidia network card.

    Also for grins we tried rh8.0 and it does not have the problem that rh9
    does.
    >
    > clg
    >



  7. Re: SLOW sockets under Redhat 9.0

    > No it's not the nvidia network card.
    >
    > Also for grins we tried rh8.0 and it does not have the problem that rh9
    > does.
    >
    >>
    >> clg
    >>

    >


    Well I'll relate my NVIDIA experience anyway and maybe it will help you.
    Lots of people go from RH8 to RH9 and find their NVIDIA networks
    painfully slow. THis is because nvidia did not make their older drivers
    compatible (build wise) with RH9. NVIDIA only made their latest driver
    RH9 compatible. So people usually think they did not change anything
    when they upgraded, but in fact they were forced to use a newer later
    NVIDIA driver. That later driver is simply garbage as far as network is
    concerned.

    The interrupting is appearantly messed up, and the driver either keeps
    interrupting network traffic, or what I think, the driver keeps
    rebooting the network and turning it completely on and off. This
    activity forces other applications to redo certain things. As a result
    you will see other apps taking up lots of processor time if you look in
    'top' apps such as nmbd and cups start to go crazy. boot times get
    dragged out.

    my solution was to fall back to an older network driver.

    have you verified that your driver is indeed that same you used on RH8?


    clg


  8. Re: SLOW sockets under Redhat 9.0

    This problem might be as simple as:

    su
    /sbin/mii-tool

    Are you at your highest capacity on your Wireless, Ethernet, or Fiber,
    including full duplex if you have that capability on your card and
    hub? If not, then do "man mii-tool" or "mii-tool --help". For
    instance, on me, I had to do...

    /sbin/mii-tool -F 100baseTx-FD

    ....in order to get full duplex capability, 100 Base T, because my card
    defaulted to half duplex. I have to do this every time I reboot my RH9
    because I don't know yet whether I can place it in a startup Bash file
    or which one.

    For some reason, this is a relatively little-known process one has to
    do with net cards in Linux. It does not appear to be a control-panel
    option in RH9.

    Also be leery of any net card logging (such as "/sbin/mii-tool -l")
    you might have enabled -- this slows down net comm tremendously. If
    you are doing wireless, I would imagine that the "Wireless Link
    Monitor" applet that you can add to the GNOME control panel might also
    be a speed impediment, but am not certain.

  9. Re: SLOW sockets under Redhat 9.0

    Google Mike wrote:
    > This problem might be as simple as:
    >
    > su
    > /sbin/mii-tool
    >
    > Are you at your highest capacity on your Wireless, Ethernet, or Fiber,
    > including full duplex if you have that capability on your card and
    > hub? If not, then do "man mii-tool" or "mii-tool --help". For
    > instance, on me, I had to do...
    >
    > /sbin/mii-tool -F 100baseTx-FD
    >
    > ...in order to get full duplex capability, 100 Base T, because my card
    > defaulted to half duplex. I have to do this every time I reboot my RH9
    > because I don't know yet whether I can place it in a startup Bash file
    > or which one.
    >
    > For some reason, this is a relatively little-known process one has to
    > do with net cards in Linux. It does not appear to be a control-panel
    > option in RH9.
    >
    > Also be leery of any net card logging (such as "/sbin/mii-tool -l")
    > you might have enabled -- this slows down net comm tremendously. If
    > you are doing wireless, I would imagine that the "Wireless Link
    > Monitor" applet that you can add to the GNOME control panel might also
    > be a speed impediment, but am not certain.


    my links autonegotiate to 100Mbps full duples.

    --
    Respectfully,


    CL Gilbert

    "Verily, verily, I say unto you, He that entereth not by the door() into
    the sheepfold{}, but climbeth up some other *way, the same is a thief
    and a robber." John 10:1

    GnuPG Key Fingerprint:
    82A6 8893 C2A1 F64E A9AD 19AE 55B2 4CD7 80D2 0A2D

    For a free Java interface to Freechess.org see
    http://www.rigidsoftware.com/Chess/chess.html


+ Reply to Thread