Return to FORMATS CVS log

Up to [ELWIX - Embedded LightWeight unIX -] / embedaddon / mtr

Version 0.95

The "split" format is for a separating the gui from the main program.
The main program can be installed setuid, and you don't want to link a
gui-library with a setuid program.


The split format is:

<pos> <host> <loss%> <rcvd pckts> <sent pckts> <best> <avg> <worst>


The "raw" format is:

hostline|xmitline|pingline|dnsline|timestampline|mplsline

hostline:
h <pos> <host IP>

xmitline:
x <pos> <seqnum>

pingline:
p <pos> <pingtime (ms)> <seqnum>

dnsline:
d <pos> <hostname>

timestampline:
t <pos> <pingtime> <timestamp>

mplsline:
m <pos> <label> <traffic_class> <bottom_stack> <ttl>

Timestampline is not  yet implemented. Need to find out how to do
ICMP timestamping first. :-)


Someone suggested to put the following text here. As to context: Some
people are wondering why mtr sometimes reports hosts beyond the
destination host.


The FINAL host will occasionally be mentioned at position n, n+1, n+2
etc.

You know traceroute, right? It sends a packet, waits for the reply to
come back and when it comes back, it sends the next packet.

If say hosts 5-8 do not send "time exceeded" packets, you'll wait a
4*3 = twelve seconds extra before you get any results on hosts 9 and
further. MTR doesn't work like that.

In theory we could send out a probe for host 1-40 all at once. But
this would pose an unnecessary burden on the network. So what we do,
is we send out probes for a max of 5 hosts beyond where we've seen a
reply. So in the example above, we'd see a reply from router at
position 4, then we'd send out 5-9 (and because the max-host is now at
9, we'll send them out at 1s/9 = 111ms intervals). When the reply from
host 9 comes back, we'll start probing for host 10-15 (at about 60ms
intervals). But suppose the network delay up to host 9 is already 200ms
and suppose our destination host is at position 11. Then by the time
the packet from host 11 comes back, we'll already have sent probe
packets for position 12, 13, and 14! Those will come back as
"destination reached" and be reported by the "raw" mode.

Curses mode will stop showing hosts with position numbers beyond the
first reply of the destination host. It could gather the information
about replies to packets sent as probes FURTHER than it actually is
into the line displayed at its true position, but it doesn't (yet).

In fact the above example is almost completely true:

% mtr -r -n -c 2 152.179.99.218 | tail -5
 13.|-- 144.232.18.238 0.0% 2 94.8 95.4 94.8 96.0 0.8
 14.|-- 152.63.16.182 0.0% 2 95.1 95.5 95.1 95.8 0.5
 15.|-- 152.63.64.106 0.0% 2 163.9 163.9 163.9 164.0 0.1
 16.|-- 152.63.50.89 50.0% 2 163.7 163.7 163.7 163.7 0.0
 17.|-- 152.179.99.218 50.0% 2 168.2 168.2 168.2 168.2 0.0
% mtr -l -c 2 152.179.99.218 | grep -v "^[dp]" |tail -7
h 10 144.232.1.41
h 11 144.232.4.96
h 16 152.179.99.218
h 17 152.179.99.218
h 18 152.179.99.218
h 12 144.232.18.238
h 13 152.63.16.182

As you can see we get the reply from the destination host at position
16 AFTER we've sent probes for position 17 and 18. When those come
back, they are reported. That's what raw mode does. It reports the raw
information.

If you write a backend for the raw mode, it's up to you to
filter/display the results.

h 10 144.232.1.41
h 11 144.232.4.96
h 12 144.232.18.238
h 13 152.63.16.182
h 14 152.63.64.106
h 15 152.63.50.89
h 16 152.179.99.218
h 17 152.179.99.218
h 18 152.179.99.218