tcp_time_wait_interval

[Stuart Biggar]

I just installed ENVI 3.6 (from Kodak) on a 280R.
The machine is running the latest version of 2.8
and is fairly up-to-date with patches.

The license manager (FLEXlm v8.1b) log file suggests
changing tcp_time_wait_interval to 2400 (from
the current default of 240000).

Are there any adverse effects from resetting
this parameter? I would probably put an ndd
command in the /etc/init.d script that starts
the license manager to change it.

Thanks,

Stuart

[Rick Jones]

Stuart Biggar <Stuart.Biggar@opt-sci.arizona.edu> wrote:

> I just installed ENVI 3.6 (from Kodak) on a 280R. The machine is
> running the latest version of 2.8 and is fairly up-to-date with
> patches.

> The license manager (FLEXlm v8.1b) log file suggests changing
> tcp_time_wait_interval to 2400 (from the current default of 240000).

> Are there any adverse effects from resetting this parameter? I
> would probably put an ndd command in the /etc/init.d script that
> starts the license manager to change it.

Only 2.4 seconds? Does the license software do something unfortunate
like use the same port number for the client side each time or
something?

TIME_WAIT is an integral part of TCP's correctness algorithms. A TCP
connection is identified by the four-tuple of local/remote IP and
local/remote port. TIME_WAIT is there to ensure that any old segments
for a given connection have timed-out in the network before a new
connection of the same name is established. It is supposed to last
for twice the maximum segment lifetime.

One common excuse for making TIME_WAIT very short is broken server
application code that does not set SO_REUSEADDR - attempts to restart
such server code will fail while there are still connections in
various states. TIME_WAIT is one of those states, but not the only
one. ESTABLISHED would also be one of those states.

rick jones
--
firebug n, the idiot who tosses a lit cigarette out his car window
these opinions are mine, all mine; HP might not want them anyway... :)
feel free to post, OR email to raj in cup.hp.com but NOT BOTH...

[Stuart Biggar]

Rick Jones wrote:

> Stuart Biggar <Stuart.Biggar@opt-sci.arizona.edu> wrote:
>

>>I just installed ENVI 3.6 (from Kodak) on a 280R. The machine is
>>running the latest version of 2.8 and is fairly up-to-date with
>>patches.

>
>

>>The license manager (FLEXlm v8.1b) log file suggests changing
>>tcp_time_wait_interval to 2400 (from the current default of 240000).

>
>

>>Are there any adverse effects from resetting this parameter? I
>>would probably put an ndd command in the /etc/init.d script that
>>starts the license manager to change it.

>
>
> Only 2.4 seconds? Does the license software do something unfortunate
> like use the same port number for the client side each time or
> something?

Possibly but I don't know at all - the following is cut from the
lmgrd log file (machine restarted earlier this AM to ensure that
lmgrd restarted automatically on reboot):

4:27:56 (lmgrd) Solaris Note:
4:27:56 (lmgrd)
4:27:56 (lmgrd) We recommend adding the following command to the boot
4:27:56 (lmgrd) scripts due to a bug in the Solaris operating system:
4:27:56 (lmgrd)
4:27:56 (lmgrd) Solaris 2.1-2.6:
4:27:56 (lmgrd) /usr/sbin/ndd -set /dev/tcp tcp_close_wait_interval 2400
4:27:56 (lmgrd) Solaris 2.7 and higher:
4:27:56 (lmgrd) /usr/sbin/ndd -set /dev/tcp tcp_time_wait_interval 2400
4:27:56 (lmgrd)
4:27:56 (lmgrd) By default on Solaris, upon stopping a license server,
4:27:56 (lmgrd) 1 to 5 minutes are required for the port to free up so it
4:27:56 (lmgrd) will restart, which can result in checkout failures.
4:27:56 (lmgrd) The command above resets this default to 2.4 seconds

Note the comment "due to a bug in the Solaris ..." - or is it a bug
in lmgrd?

>
> TIME_WAIT is an integral part of TCP's correctness algorithms. A TCP
> connection is identified by the four-tuple of local/remote IP and
> local/remote port. TIME_WAIT is there to ensure that any old segments
> for a given connection have timed-out in the network before a new
> connection of the same name is established. It is supposed to last
> for twice the maximum segment lifetime.
>
> One common excuse for making TIME_WAIT very short is broken server
> application code that does not set SO_REUSEADDR - attempts to restart
> such server code will fail while there are still connections in
> various states. TIME_WAIT is one of those states, but not the only
> one. ESTABLISHED would also be one of those states.

I expect I will not change the parameter as we shouldn't really
have to stop the license server. Thanks for the information.

Stuart

>
> rick jones

[Barry Margolin]

In article <bgo898$a5a$1@oasis.ccit.arizona.edu>,
Stuart Biggar <Stuart.Biggar@opt-sci.arizona.edu> wrote:

> 4:27:56 (lmgrd) Solaris Note:
> 4:27:56 (lmgrd)
> 4:27:56 (lmgrd) We recommend adding the following command to the boot
> 4:27:56 (lmgrd) scripts due to a bug in the Solaris operating system:
> 4:27:56 (lmgrd)
> 4:27:56 (lmgrd) Solaris 2.1-2.6:
> 4:27:56 (lmgrd) /usr/sbin/ndd -set /dev/tcp tcp_close_wait_interval 2400
> 4:27:56 (lmgrd) Solaris 2.7 and higher:
> 4:27:56 (lmgrd) /usr/sbin/ndd -set /dev/tcp tcp_time_wait_interval 2400
> 4:27:56 (lmgrd)
> 4:27:56 (lmgrd) By default on Solaris, upon stopping a license server,
> 4:27:56 (lmgrd) 1 to 5 minutes are required for the port to free up so it
> 4:27:56 (lmgrd) will restart, which can result in checkout failures.
> 4:27:56 (lmgrd) The command above resets this default to 2.4 seconds
>
>Note the comment "due to a bug in the Solaris ..." - or is it a bug
>in lmgrd?

I initially thought it sounded like a bug in their code, because simply
using the SO_REUSEADDR socket option should allow the port to be reused.

But the fact that they single out Solaris for this recommendation suggests
that they already do this. If they didn't, they should have the same
problem on other platforms as well. So maybe there's some other problem,
although I've never heard about it. Considering how popular Solaris is for
servers, I'd expect this to be a more well-known problem.

--
Barry Margolin, [email]barry.margolin@level3.com[/email]
Level(3), Woburn, MA
*** DON'T SEND TECHNICAL QUESTIONS DIRECTLY TO ME, post them to newsgroups.
Please DON'T copy followups to me -- I'll assume it wasn't posted to the group.

[Oscar del Rio]

> The license manager (FLEXlm v8.1b) log file suggests

> changing tcp_time_wait_interval to 2400 (from
> the current default of 240000).

[url]http://docs.sun.com/db/doc/806-7009/[/url]
TCP/IP Tunable Parameters

tcp_time_wait_interval

Description
The time in milliseconds a TCP connection stays in TIME-WAIT state.
For more information, refer to RFC 1122, 4.2.2.13.
Default
60,000 (60 seconds)
Range
1 second to 10 minutes
Dynamic?
Yes
When to Change
Do not set the value lower than 60 seconds.
For more information, refer to RFC 1122, 4.2.2.13.


Some versions of FlexLM insisted in changing this in Solaris to 2400.
The manual for the current version (9.2 Jul 2003) seems to have that
recommendation removed; I do not know if lmgrd 9.2 still logs
the warning on startup; 9.0 did:

(lmgrd) Solaris Note:
(lmgrd)
(lmgrd) We recommend adding the following command to the boot
(lmgrd) scripts due to a bug in the Solaris operating system:
(lmgrd)
(lmgrd) Solaris 2.1-2.6:
(lmgrd) /usr/sbin/ndd -set /dev/tcp tcp_close_wait_interval 2400
(lmgrd) Solaris 2.7 and higher:
(lmgrd) /usr/sbin/ndd -set /dev/tcp tcp_time_wait_interval 2400
(lmgrd)
(lmgrd) By default on Solaris, upon stopping a license server,
(lmgrd) 1 to 5 minutes are required for the port to free up so it
(lmgrd) will restart, which can result in checkout failures.
(lmgrd) The command above resets this default to 2.4 seconds
(lmgrd) -----------------------------------------------
(lmgrd) FLEXlm (v9.0 ) started on hostname (Sun) (7/29/2003)
(lmgrd) Copyright (c) 1988-2003 by Macrovision Corporation.

Does anyone know what's the "bug in the Solaris operating system"
they are talking about?