NAME
paris-traceroute - print the IP-level routes between two Internet
hosts.
SYNOPSIS
paris-traceroute [ -fhilnqvV ] [ -b initial_id ]
[ -d dest_port ] [ -a algorithm ] [ -f first_ttl ]
[ -L packetlen ] [ -m max_ttl ]
[ -M max_missing_hops ] [ -p protocol ]
[ -q nqueries ] [ -s source_port ] [ -t tos ]
[ -T delaymsecs ] [ -w waittime ]
host
DESCRIPTION
Paris traceroute is a new version of the well-known network diagnosis
tool. It addresses problems caused by load balancers with the initial
traceroute(8) implementation. By controlling the flow identifier of the
probes, it is able to follow accurate paths in networks with load
balancers. It is also able to find all the load balanced paths to the
destination. Finally, it enriches its output with information
extracted from the received packets, allowing a more precise analysis
of the discovered paths.
Options are:
-a Set the probing algorithm:
hopbyhop
Send q (configured with the -q flag) probes with the same TTL,
then wait for all the replies or a timeout. Increment the TTL
and reiter the operation until we reach the destination. All
packets hold the same flow identifier.
packetbypacket
It is the classic traceroute(8) algorithm: send one probe at a
time, then wait for a reply or a timeout. Reiter the operation
until we reach the destination.
concurrent
Send all the probes from min_ttl to max_ttl and wait for all
replies or a timeout.
scout Send a scout probe with a ttl max to the destination. If the
destination can be reached, compute the number of hops used to
reach the destination and start the concurrent algorithm with a
max_ttl equal to this number of hops. If the destination cannot
be reached, the hopbyhop algorithm will be used instead. This
algorithm is only usable with UDP probes.
exhaustive
Print all the possible "load balanced" paths to the destination.
(See section EXHAUSTIVE ALGORITHM )
-b Set the initial probe identifier.
-d Set the the UDP/TCP destination port (default: 33457).
-f Set the initial ttl (default: 1).
-h Print help.
-i Print the "IP Identifier" value of the responses. It is used to
identify the different interfaces of a router, or uncover NAT
boxes.
-l Display the ttl value of the reply. Useful to study asymmetric
routing and NAT boxes.
-L Set the data length to be used in outgoing packets. (default:
0).
-m Set the maximum ttl (default: 30).
-M Set the maximum number of consecutive unresponsive hops which
causes the program to abort (default 3).
-n Print hop addresses numerically. The default is to print also
hostnames.
-p Set the protocol to use (possible values: udp, tcp, icmp).
-q Set the number of probes per hop (default: 3).
-s Set the UDP/TCP source port (default: 33456).
-t Set the Type of Service (default: 0). This field is taken into
account by many per-flow load balancers: in presence of such a
load balancer, packets having different TOS values are likely to
follow a different paths.
-T Set the time to wait between probes, in milliseconds (default
50ms).
-v Print debug messages.
-V Print the program version.
-w Set the time to wait for a response, in milliseconds (default
5000ms).
EXHAUSTIVE ALGORITHM
With the deployment of load balancing, there is no longer only one path
between two Internet hosts. This algorithm sends enough probes at each
hop to find all the possible interfaces. Unlike the other algorithms,
it varies the flow identifier of the probes in a controlled manner, to
ensure the discovery of all the interfaces with a high confidence
degree. It also categorizes load balancers as "per-packet" (pseudo-
random, round-robin packet balancing) or "per-flow" (packets belonging
to the same flow follow the same path).
In case of per-flow load balancing, it prints additional information to
track flows. The following trace shows the enriched output:
14 1.1.1.1:0,1,3 539.065 ms 1.1.1.2:2,4,5 492.152 ms
15 2.2.2.2:0,1,3 563.163 ms 2.2.2.3:2,4,5 470.919 ms
Integers listed after the interface addresses are "flow identifiers":
they are used to identify a flow in the set of interfaces found by the
algorithm. For example, flow #0 traverses interfaces 1.1.1.1 and
2.2.2.2. This is the same for flows 1 and 3 while flows 2, 4 and 5
traverse 1.1.1.2 and 2.2.2.3.
OUTPUT
The following information are extracted from the response packets and
displayed:
Response TTL
The TTL of the responses (from the routers and the destination)
is optionally displayed in square brackets (Use the -l flag ).
Original TTL
This is the TTL of the probe when it was received and dropped by
the router. If the original TTL is different than 1, it is
displayed with a !Tx symbol, where x is the value of the TTL.
For example, !T0 indicates that the value of the TTL was 0 when
the probe reached the router that discarded it.
IP Identifier
This the identifier of the IP error packet sent by the router.
This field is set with the value of an internal 16-bit counter
usually incremented for each packet sent. This value is
optionally displayed inside brackets. For instance {1234}
indicates that the probe had its identifier set to 1234.
MPLS labels
If the packet contains ICMP extensions for MPLS, the MPLS label
stack is diplayed in an additionnal line just after the current
hop line. Labels of the same stack are separated with a "|"
character.
Other ICMP error messages
Paris traceroutes uses the same convensions as traceroute(8) to
display unexpected ICMP messages (i.e. different than
TIME_EXCEEDED, PORT_UNREACHABLE and ECHO_REPLY).
SEE ALSO
traceroute(8), pathchar(8), netstat(1), ping(8)
AUTHOR
The initial version of traceroute(8) was implemented by Van Jacobson
from a suggestion by Steve Deering. Paris traceroute was implemented
by Xavier Cuvellier. Debugged and enhanced by Brice Augustin.
The current version is available at:
http://www.paris-traceroute.net
BUGS
Please send bug reports to brice.augustin@rp.lip6.fr.