What Causes TCP Send Queue to Grow ("netstat" -an output)?

Hi,

I keep seeing the Send Queue Column growing from "netstat -an", can
someone tell me what causes this to happen? I believe in RH Linux that
the buffer is at 128K, but I'm not sure. We have been seeing some
problems lately with performance, but am not sure how to pinpoint
this.

The column with the 53576 number seems very problematic...

kernel - 2.4.20 on RH Linux 7.3

-----------------------------------------------

Active Internet connections (servers and established)
tcp 0 53576 172.22.98.99:34930 172.27.23.99:8400


Thanks!

 

What we have hear is a failure to communicate. It means that either
172.27.23.99 is not accepting data right now, or there is some problem
between the two (your nic, the other nic, or network/routing between
them). Recv-Q and Send-Q should normally be near zero.

If these are dummy IPs and actually connects over the internet, it is not
uncommon to have occasional router interruptions or looping (traceroute
the other IP when that happens to see if that gives a clue).

 

The buffer may be 128k, but linux assumes half the buffer is used for
kernal structures so you only have 64k of space.

Terry

 

How would I go about increasing the buffer size then? Does that mean
it's 128K buffer in total for sending/receiving? I may need to
increase these as we have been having some issues in the office.

 

You can use setsockopt() to change the size of the send/receive buffers
within an application and sysctl to change default system settings for
all applications. See the man page socket(7) which gives details
on what values to change.