Additional TCP_INFO items

[GoogleCodeExporter]

Explaination of new feature

Currently we are using TCP_INFO to display TCP retransmission information.

There are other fields we might want to consider including as well. Perhaps in 
"verbose mode".

Here is the full list of fields:

            tcp_info.tcpi_last_data_sent,
            tcp_info.tcpi_last_data_recv,
            tcp_info.tcpi_snd_cwnd,
            tcp_info.tcpi_snd_ssthresh,
            tcp_info.tcpi_rcv_ssthresh,
            tcp_info.tcpi_rtt,
            tcp_info.tcpi_rttvar,
            tcp_info.tcpi_unacked,
            tcp_info.tcpi_sacked,
            tcp_info.tcpi_lost,
            tcp_info.tcpi_retrans,
            tcp_info.tcpi_fackets

I suggest we add the following:
            tcp_info.tcpi_snd_cwnd,
            tcp_info.tcpi_sacked,

Anything else folks would like added?

Original issue reported on code.google.com by bltier...@es.net on 4 Nov 2013 at 11:06

[GoogleCodeExporter]

I think these are enabled in linux  by  looking at the code

#if defined(linux)
    sprintf(message, report_tcpInfo, r->tcpInfo.tcpi_snd_cwnd, r->tcpInfo.tcpi_snd_ssthresh,
            r->tcpInfo.tcpi_rcv_ssthresh, r->tcpInfo.tcpi_unacked, r->tcpInfo.tcpi_sacked,
            r->tcpInfo.tcpi_lost, r->tcpInfo.tcpi_retrans, r->tcpInfo.tcpi_fackets,
            r->tcpInfo.tcpi_rtt, r->tcpInfo.tcpi_reordering);
#endif

Original comment by susant.sahani on 30 Nov 2013 at 6:23

[GoogleCodeExporter]

That code is currently not being called.

Original comment by jef.posk...@gmail.com on 30 Nov 2013 at 7:50

[GoogleCodeExporter]

Grab.

Original comment by bmah@es.net on 3 Dec 2013 at 12:47

• Changed state: Accepted

[GoogleCodeExporter]

Committed some WIP in 4cfce137e89c.

Outstanding issues:

We haven't decided if or what quantities should go under a --verbose flag.  We 
could dump a lot more stuff unconditionally into the --json output.

No FreeBSD support.  I think it's possible to pull some similar quantities from 
FreeBSD's version of tcp_info, which isn't *quite* like the Linux version.

Representation of snd_cwnd...this quantity appears to be expressed in segments, 
but putting it in octets might be more useful?

Summarization of snd_cwnd; should we show an average, maximum, or...???..

We might modify (or even defer) this implementation before the upcoming 3.0 
release.

Original comment by bmah@es.net on 18 Dec 2013 at 11:14

[GoogleCodeExporter]

Revision 5a0e6b6456cf took out the tcpi_sacked support...it didn't really keep 
a cumulative count like we thought.  snd_cwnd is now expressed in octets.

616ac79bd80b suppresses the snd_cwnd output in the human-readable output 
pending resolution of some issues about formatting and representation of these 
values.

Original comment by bmah@es.net on 19 Dec 2013 at 10:48

[GoogleCodeExporter]

This issue was closed by revision c83a33701051.

Original comment by bmah@es.net on 14 Feb 2014 at 7:18

• Changed state: Fixed

[GoogleCodeExporter]

Revision c83a33701051 brings back snd_cwnd output in a human-friendly form, and 
also fixes up some headers so they make sense.

Resolving as Fixed.

Original comment by bmah@es.net on 14 Feb 2014 at 7:19