Keeping track of TCP retries.

I'm trying to figure out if it's possible to track or increment a
counter that will keep track of the number of times the OS is forced to
"retry" a TCP packet. In our case, this will help us identify customer
connections which are suffering from connection problems without having
to run an icmp ping to them at all times.

I'm a little foggy on the tcpdump output, but it appears that the
sequence numbers stay the same when a TCP connection times out and
needs to be retried, but I'm not sure of a great way to identify and
track this stuff, or if it's even smart to do given lots of network
traffic.

Any ideas or suggestions would be appreciated.

(E-Mail Removed) wrote:
> I'm trying to figure out if it's possible to track or increment a
> counter that will keep track of the number of times the OS is forced
> to "retry" a TCP packet. In our case, this will help us identify
> customer connections which are suffering from connection problems
> without having to run an icmp ping to them at all times.

> I'm a little foggy on the tcpdump output, but it appears that the
> sequence numbers stay the same when a TCP connection times out and
> needs to be retried, but I'm not sure of a great way to identify and
> track this stuff, or if it's even smart to do given lots of network
> traffic.

Which is it - retransmission of TCP segments (packet) as you imply in
the first paragraph, or reconnection after TCP connections timeout as
implied in the second paragraph?

A retransmitted TCP segment will overlap in sequence space with
segments already sent. In tcpdump you will see the sequence number go
back, or remain the same (if there was only one segment outstanding on
the connection and it was the one lost).

On a 2.6 kernel at least, the output of /proc/net/tcp looks promising.

rick jones
--
denial, anger, bargaining, depression, acceptance, rebirth...
where do you want to be today?
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...

In comp.os.linux.networking (E-Mail Removed):
> I'm trying to figure out if it's possible to track or increment a
> counter that will keep track of the number of times the OS is forced to
> "retry" a TCP packet. In our case, this will help us identify customer
> connections which are suffering from connection problems without having
> to run an icmp ping to them at all times.

netstat -s

Might help.

Hint:
man netstat

[..]

Good luck

--
Michael Heiming (X-PGP-Sig > GPG-Key ID: EDD27B94)
mail: echo (E-Mail Removed) | perl -pe 'y/a-z/n-za-m/'
#bofh excuse 157: Incorrect time synchronization