Summary: Solaris 2.4 Network kernal params...

From: Joe Dietz (
Date: Thu Apr 20 1995 - 03:25:18 CDT

Subject: Solaris 2.3 Network Kernal params...?

Could someone point me to the kernal parameters that would change the "ACK"
timeout time for TCP/IP...

We are having some network problems...The Solaris system is resending data
because it does not see an ACK from the client in time...we know the data is
getting to the client but not on the first try...the application works but
the performance is in the

While we are fixing our network/router problems I would like to extend the
amount of time we will wait for the ACK....this should cut down on the number
of retries...and hopefully relieve some of the network load...

Any other solutions or recommandations are welcome...

Bravo! To all who pitched in....

I have changed the following two parameters and my retransmision are gone!

ndd /dev/tcp tcp_rexmit_interval_min
ndd /dev/tcp tcp_deferred_ack_interval

I shot from the hip and doubled the two parameters....;-)

I have verification from traces on network sniffers that the are no
retransmitions. I figured it would help but did not expect the retransmisions
to go away...

I'm not sure if we needed to change both parameters...more testing will

If anyone has electronic documentation on the tcp/ip/udp parameters I am
interested...I have full docs and there is nothing on these parameters....

Thanks again...


From: (Amy Hollander)

Please summarize or send me the result of your quest.

From: "Srinivasa R. Yalavarthy" <>

        type "ndd /dev/tcp \?" will give you all tunable parameters. choose the
        parameter that you wanna change type the following

        ndd -set /dev/tcp <parameter> <value>

From: (tommy hill)

If you receive any replies, I would be very interested in what
is sent. I am having similiar problems with a Web server on a
SPARCserver. When I do a netstat -an, the number of
TIME_WAIT and CLOSE_WAIT's is in the hundreds and I really
think that there is a problem with Solaris 2.3 and TCP/IP as
far as thresholds and/or limits.
I have experimented with the ndd command and reset the number
of tcp_conn_req_max to the value 28, hoping that it would help.
There has been a little improvement but I would really like
to take care of all of the WAIT's that are appearing.

From: (Koen Peeters)

Is your application using TCP or UDP ??
For UDP you can not control the retransmit, it is application dependent.

these are the default values for retransmit for TCP

# ndd /dev/tcp tcp_rexmit_interval_initial
# ndd /dev/tcp tcp_rexmit_interval_max
# ndd /dev/tcp tcp_rexmit_interval_min

you can change them with

# ndd -set /dev/tcp tcp_rexmit_interval_initial 500

The values are not very well documented so you'll have experiment a bit.

From: (Glenn Satchell - Uniq Professional Services)

Check out 'ndd /dev/tcp ?' which will print out a list of the variables
that you can change for the TCP module. Likewise for /dev/udp for the
UDP module. Put this string in one of the initialisation files, such as

From: (Ruud van Poelgeest)

I don't know the kernel parameters, but there is a command 'ndd' which has a lot parameters for ip, tcp and udp. I hope that helps you.

Do a ndd /dev/ip ? for ip parameters,
   a ndd /dev/tcp ? for tcp parameters
 and ndd /dev/udp ? for udp parameters

This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:10:22 CDT