Idea/Suggestion:
Take a look at DOS CMD / Unix CMD "netstat -an" output on the Server.
If you see a significant amount of sockets in "TIME_WAIT" state, you may need
to add the "TcpTimedWaitDelay" parameter to the Registry, (for Windows). There
are ways to modify this in various Unix platforms - you'll need to dig out the
particulars for your 'nix variant.

A lot of applications will call/open a socket, send data, then close the
socket, rather than re-use, and then call a new socket next time. When next
connection is opened, or a lot of connections are opened, perhaps the Server
does not have the "free resources" to clean up these connection/sockets that
are lingering around, as the OS is designed to perform such clean up in the
background when it has the capacity to do so. Busy Servers may not be able to
free up these resources as fast as they're being called/used. (This behavior
exists across all Windows and Unix platforms).

For Windows, you can add a Registry setting called "TcpTimedWaitDelay" which
does not exist by default, but it does use a default parameter, (240 I
believe). which may be way too long for many/most applications, especially if
applications do not "re-use" alread opened socket/ports.  By adding this
parameter as specificed in the MS TechNote, and setting its value to "30"
decimal (i.e. "30 seconds"), this will help free up significant amounts of
ports stuck in "TIME_WAIT" state in a significantly faster manner, proving in
more smooth server & network operations and sockets availabilty.

http://technet2.microsoft.com/windowsserver/en/library/38b8bf76-b7d3-473c-84e
8-e657c0c619d11033.mspx?mfr=true

I have seen this behavior numerous times, across numerous applications and
across numerous platforms, verticals, etc...

I hope this helps.

Regards,
Scott Richardson
Senior Systems Engineer / Consultant
Sr. Product Support Engineer
Marlborough, MA 01752
Email: CheetahFTL    at      comcast.net
****************************************
DPMonitor - http://www.deltek.us


----- Original Message -----
From: "Kevin King" <[EMAIL PROTECTED]>
To: <[email protected]>
Sent: Wednesday, May 28, 2008 12:07 PM
Subject: Re: [U2] Re: ODBC Connection Hanging Around


> Well, the CLOSE appears to have been a red herring.  Everything ran fine
for
> several hours, then all at once a half dozen sessions didn't get
> disconnected.  And again it appears to have been when there were a lot of
> simultaneous sessions.  I'm leaning more and more to a threading issue with
> the Unidata ODBC driver.
>
> On Wed, May 28, 2008 at 8:15 AM, Kevin King <[EMAIL PROTECTED]> wrote:
>
>> The client terminates fine.  It's the server session that doesn't
>> disconnect.  And in a weird twist that I'm still watching, I added a
generic
>> CLOSE statement to close all the files being used by the subroutine before
>> it returns to the caller and since making that change the server has not
>> left any connections open.  But like I said, I'm still watching this to
see
>> if this was the silver bullet or merely a coincidence.
-------
u2-users mailing list
[email protected]
To unsubscribe please visit http://listserver.u2ug.org/

Reply via email to