readme.mtrace

mtrace源码

README.mtrace,v 5.2 1998/12/04 05:04:15 fenner Exp

	Multicast Traceroute

	Release 5.2
	December 3, 1998

	available from ftp.parc.xerox.com,
	file pub/net-research/ipmulti/mtrace-5.2.tar.gz
	binaries pub/net-research/ipmulti/mtrace-5.2-sparc-sunos41x.tar.gz
	         pub/net-research/ipmulti/mtrace-5.2-sparc-solaris2.tar.gz
	         pub/net-research/ipmulti/mtrace-5.2-alpha-osf1.tar.gz
	         pub/net-research/ipmulti/mtrace-5.2-sgi-irix.tar.gz


The 5.2 release fixes the following bugs:

o 100% loss was reported as "?/nn = --"; switch it back to 100%

o TTL calculation didn't handle IOS's claim that packets with TTL 0 would
  be forwarded.

o Part of the byte-swapping heuristic was removed; it caused more problems
  than it solved.

o Hop names are now printed correctly in passive mode

o The interface list is properly retrieved from the kernel; selecting
  an interface now works correctly on multi-homed 4.3-Tahoe-and-later
  hosts.  It properly skips down interfaces, thanks to John Kay,
  and stops when it finds the right interface, thanks to Mark Foster.

o mtrace no longer depends on trying to figure out how the compiler is
  going to pack an 8-bit and 24-bit integer into a 32-bit value and
  instead explicitly extracts the values from a 32-bit value (tr_rttlqid).

o passive mtrace joins the group on the interface specified with -i
  and now always prints * * * for the query source since it's unknown
  (thanks to Dave Thaler)

The 5.2 release has the following new features:

o New option -L <n> to only print stats when loss is greater than n%

o New option -f <n> to start a hop-by-hop trace at hop n

o Workaround for IOS bug CSCdi68628 which causes incorrect routers to
  reply to 3rd-party mtrace requests

o Options RAW_INPUT_IS_RAW and RAW_OUTPUT_IS_RAW for compilation on
  Linux and OpenBSD

o New routing protocol codes

o Passive mode now ignores most duplicate responses (instead of treating
  them as individual responses and printing out multiple stats runs
  at once)

o Workaround for some routers returning integer seconds instead of
  in units of 1/65535th as the spec says

o The default group of 224.2.0.1 has been removed; the default is now
  a "weak" mtrace (i.e. group 0.0.0.0).  This can cause problems with
  older versions of IOS; if you see "No route" where it doesn't seem
  to make sense, try the trace again specifying a group.



------
Release 5.1
December 19, 1996


The 5.1 release fixes the following bugs:

o Workaround for "0 pps" printed for little-endian hops.  There is a
  missing parenthesis in mrouted 3.8's mtrace responder, causing half
  of a calculation to be byte-swapped and half not.  mtrace now uses
  the time the response was received instead of the timestamp in the
  packet if it detects this bug.

o TTL required is now calculated correctly.

o Wrong-type packets are properly ignored.  If multiple mtrace's are
  running on a single machine, it's possible that a DVMRP_NEIGHBORS2
  ("mrinfo") packet could have been misinterpreted as an mtrace packet,
  displaying a long trace with lots of "0.0.0.0" entries.

o "-m" traces that take less hops than requested are now handled properly.

o No longer allows magic address expansion for multicast addresses, so
  "224.100.102" is now a syntax error instead of a synonym for "224.100.102.0".

o If statistics were aborted because the route changed, don't count them
  as having been printed.

o If the last hop forgot to fill in its IP address, fill it in with the
  source address of the returned packet.

o Missing statistics detection is now more robust; it will just print
  blank lines rather than the confusing "58/58=100%" then "-58/0 = --%".

The 5.1 release has the following new features:

o Netmask printing (with -v) and default route notification

o Traces are no longer restarted from scratch when the route changes;
  they are continued from the information already known.

o The new Watch-Path mode (-P) allows an individual path to be monitored
  periodically; mtrace will print the path initially and only when it
  changes.

o Default multicast TTL changed from 64 to 127, to better match reality.

o Per-hop statistics are normally only printed as a packet-per-second count;
  there are so many bugs in the collection of these counters that the packet
  loss that they report is generally misleading.  The -T flag may be used
  to include these numbers anyway.

o Don't print statistics if there is no group specified, since the per-group
  statistics are the only ones that are meaningful.

o Change the default destination when using "-g <gwy>" to <gwy> if it is
  a different subnet than the host running mtrace.  This elminiates the need
  to use "mtrace -g foo bar foo".

o Print "[reset counters]" on a hop that has reset its counters.

o Allow the number of extra hops to be traced after a non-responding
  router to be specified with the "-e <nhops>" flag.

o Statistics collection in passive mode (slightly buggy)

o The -O flag may be used to supress use of the Router Alert IP option.
  Some versions of Cisco's IOS corrupt multicast traceroute replies when
  the request is sent with IP options, so if the last hop is a Cisco you
  may need to use -O.