Network Working Group                                        D. L. Mills
Request for Comments: 981                               M/A-COM Linkabit
                                                              March 1986

            An Experimental Multiple-Path Routing Algorithm

Status of This Memo

   This RFC describes an experimental, multiple-path routing algorithm
   designed for a packet-radio broadcast channel and discusses the
   design and testing of a prototype implementation.  It is presented as
   an example of a class of routing algorithms and data-base management
   techniques that may find wider application in the Internet community.
   Of particular interest may be the mechanisms to compute, select and
   rank a potentially large number of speculative routes with respect to
   the limited cumputational resources available.  Discussion and
   suggestions for improvements are welcomed.  Distribution of this memo
   is unlimited.

Abstract

   This document introduces wiretap algorithms, which are a class of
   routing algorithms that compute quasi-optimum routes for stations
   sharing a broadcast channel, but with some stations hidden from
   others. The wiretapper observes the paths (source routes) used by
   other stations sending traffic on the channel and, using a heuristic
   set of factors and weights, constructs speculative paths for its own
   traffic.  A prototype algorithm, called here the Wiretap Algorithm,
   has been designed for the AX.25 packet-radio channel.  Its design is
   similar in many respects to the shortest-path-first (spf) algorithm
   used in the ARPANET and elsewhere, and is in fact a variation in the
   class of algorithms, including the Viterbi Algorithm, that construct
   optimum paths on a graph according to a distance computed as a
   weighted sum of factors assigned to the nodes and edges.

   The Wiretap Algorithm differs from conventional algorithms in that it
   computes not only the primary route (a minimum-distance path), but
   also additional paths ordered by distance, which serve as alternate
   routes should the primary route fail.  This feature is also useful
   for the discovery of new paths not previously observed on the
   channel.

   Since the amateur AX.25 packet-radio channel is very active in the
   Washington, DC, area and carries a good deal of traffic under
   punishing conditions, it was considered a sufficiently heroic
   environment for a convincing demonstration of the prototype
   algorithm.  It was implemented as part of an IP/TCP driver for the
   LSI-11 processor running the "fuzzball" operating system.  The driver
   is connected via serial line to a 6809-based TAPR-1 processor running
   the WA8DED firmware, which controls the radio equipmnet in both

Mills                                                           [Page 1]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

   virtual-circuit and datagram modes. The prototype implementation
   provides primary and alternate routes, can route around congested
   areas and can change routes during a connection. This document
   describes the design, implementation and initial testing of the
   algorithm.

1.  Introduction

   This document describes the design, implementation and initial
   testing of the Wiretap Algorithm, a dynamic routing algorithm for the
   AX.25 packet-radio channel [4].  The AX.25 channel operates in CSMA
   contention mode at VHF frequencies using AFSK/FM modulation at 1200
   bps. The AX.25 protocol itself is similar to X.25 link-layer protocol
   LAPB, but with an extended frame header consisting of a string of
   radio callsigns representing a path, usually selected by the
   operator, between two end stations, possibly via one or more
   intermediate packet repeaters or digipeaters.  Most stations can
   operate simultaneously as intermediate systems digipeaters) and as
   end systems with respect to the ISO model.

   Wiretap uses passive monitoring of frames transmitted on the channel
   in order to build a dynamic data base which can be used to determine
   optimum routes.  The algorithm operates in real time and generates a
   set of paths ordered by increasing total distance, as determined by a
   shortest-path-first procedure similar to that used now in the ARPANET
   and planned for use in the new Internet gateway system [2].  The
   implementation provides optimum routes (with respect to the factors
   and weights selected) at initial-connection time for virtual
   circuits, as well as for each datagram transmission.  This document
   is an initial status report and overview of the prototype
   implementation for the LSI-11 processor running the "fuzzball"
   operating system.

   The principal advantage in the use of routing algorithms like Wiretap
   is that digipeater paths can be avoided when direct paths are
   available, with digipeaters used only when necessary and also to
   discover hidden stations.  In the present exploratory stage of
   evolution, the scope of Wiretap has been intentionally restricted to
   passive monitoring.  In a later stage the scope may be extended to
   include the use of active probes to discover hidden stations and the
   use of clustering techniques to manage the distribution of large
   quantities of routing information.

   The AX.25 channel interface is the 6809-based TAPR-1 processor
   running the WA8DED firmware (version 1.0) and connected to the LSI-11
   by a 4800-bps serial line.  The WA8DED firmware produces as an option
   a monitor report for each received frame of a selected type,

Mills                                                           [Page 2]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

   including U, I and S frames.  Wiretap processes each of these to
   extract routing information and (optionally) saves them in the system
   log file. Following is a typical report:

      fm KS3Q to W4CQI via WB4JFI-5* WB4APR-6 ctl I11 pid F0

   The originating station is KS3Q and the destination is W4CQI.  The
   frame has been digipeated first by WB4JFI-5 and then WB4APR-6, is an
   I frame (sequence numbers follow the I indicator) and has protocol
   identifier F0 (hex).  The asterisk "*" indicates the report was
   received from that station.  If no asterisk appears, the report was
   received from the originator.

2.  Design Principles

   A path is a concatenation of directed links originating at one
   station, extending through one or more digipeaters and terminating at
   another station.  Each link is characterized by a set of factors such
   as cost, delay or throughput that can be computed or estimated.
   Wiretap computes several intrinsic factors for each link and updates
   the routing data base, consisting of node and link tables.  The
   weighted sum of these factors for each link is the distance of that
   link, while the sum of the distances for each link in the path is the
   distance of that path.

   It is the intent of the Wiretap design that the distance of a link
   reflect the a-priori probability that a packet will successfully
   negotiate that link relative to the other choices possible at the
   sending node.  Thus, the probability of a non-looping path is the
   product of the probabilities of its links.  Following the technique
   of Viterbi [1], it is convenient to represent distance as a
   logarithmic transformation of probability, which then becomes a
   metric.  However, in the following the underlying probabilities are
   not considered directly, since the distances are estimated on a
   heuristic basis.

   Wiretap incorporates an algorithm which constructs a set of paths,
   ordered by distance, between given end stations according to the
   factors and weights contained in the routing data base.  Such paths
   can be considered optimum routes between these stations with respect
   to the given assignment of factors and weights.  In the prototype
   implementation one of the end stations must be the Wiretap station
   itself;  however, in principle, the Wiretap station can generate
   routes for other stations subject to the applicability of the
   information in its data base.

   Note that Wiretap in effect constructs minimum-distance paths in the

Mills                                                           [Page 3]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

   direction from the destination station to the Wiretap station and,
   based on that information, then computes the optimum reciprocal
   routes from the Wiretap station to the destination station.  The
   expectation is that the destination station also runs its own routing
   algorithm, which then computes its own optimum reciprocal routes
   (i.e.  the optimum direct routes from the Wiretap station).  However,
   the routing data bases at the two stations may diverge due to
   congestion or hidden stations, so that the computed routes may not
   coincide.

   In principle, Wiretap-computed routes can be fine-tuned using
   information provided not only by its directly communicating stations
   but others that may hear them as well.  The most interesting scenario
   would be for all stations to exchange Wiretap information using a
   suitable distributed protocol, but this is at the moment beyond the
   scope of the prototype implementation.  Nevertheless, suboptimum but
   useful paths can be obtained in the traditional and simple way with
   one station using a Wiretap-computed route and the other its
   reciprocal, as determined from the received frame header.  Thus,
   Wiretap is compatible with existing channel procedures and protocols.

3.  Implementation Overview

   The prototype Wiretap implementation for the LSI-11 includes two
   routines, the wiretap routine, which extracts information from
   received monitor headers and builds the routing data base, and the
   routing routine, which calculates paths using the information in the
   data base. The data base consists of three tables, the channel table,
   node table and link table.  The channel table includes an entry for
   each channel (virtual circuit) supported by the TAPR-1 processor
   running the WA8DED firmware, five in the present configuration.  The
   structure and use of this table are only incidental to the algorithm
   and will not be discussed further.

   The node table includes an entry for each distinct callsign (which
   may be a collective or beacon identifier) heard on the channel,
   together with node-related routing information, the latest computed
   route and other miscellaneous information.  The table is indexed by
   node ID (NID), which is used in the computed route and in other
   tables instead of the awkward callsign string.  The link table
   contains an entry for each distinct (unordered) node pair observed in
   a monitor header.  Each entry includes the from-NID and to-NID of the
   first instance found, together with link-related routing information
   and other miscellaneous information.  Both tables are dynamically
   managed using a cache algorithm based on a weighted
   least-recently-used replacement mechanism described later.

Mills                                                           [Page 4]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

   The example discussed in Appendix A includes candidate node and link
   tables for illustration.  These tables were constructed in real time
   by the prototype implementation from off-the-air monitor headers
   collected over a typical 24-hour period.  Each node table entry
   requires 26 bytes and each link table entry four bytes.  The maximum
   size of the node table is presently 75 entries, while that of the
   link table is 150 entries.  Once the cache algorithm has stabilized
   for a day or two, it is normal to have about 60 entries in the node
   table and 100 entries in the link table.

   The node table and link table together contain all the information
   necessary to construct a network graph, as well as calculate paths on
   that graph between any two end stations, not just those involving the
   Wiretap station.  Note, however, that the Wiretap station does not in
   general hear all other stations on the channel, so may choose
   suboptimum routes.  However, in the Washington, DC, area most
   stations use one of several digipeaters, which are in general heard
   reliably by other stations in the area.  Thus, a Wiretap station can
   eventually capture routes to almost all other stations using the
   above tables and the routing algorithm described later.

4.  The Wiretap Routine

   The wiretap routine is called to process each monitor header.  It
   extracts each callsign from the header in turn and searches the node
   table for corresponding NID, making a new entry and NID if not
   already there.  The result is a string of NIDs, starting at the
   originating station, extending through a maximum of eight digipeaters
   and ending at the destination station.  For each pair of NIDs along
   this string the link table is searched for either the direct link, as
   indicated in the string, or its reciprocal;  that is, the direction
   towards the originator.

   The operations that occur at this point can be illustrated by the
   following diagram, which represents a monitor header with apparent
   path from station 4 to station 6 via digipeaters 7, 2 and 9 in
   sequence.  It happens the header was heard by the Wiretap station (0)
   from station 2.

                   (4)     (7)     (2)     (9)     (6)
              orig o------>o<=====>o------>o------>o dest
                                   |
                                   |
                                   V
                                  (0)
                                wiretap

Mills                                                           [Page 5]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

   Presumably, the fact that the header was heard from station 2
   indicates the path from station 4 to station 2 and then to station 0
   is viable, so that each link along this path can be marked "heard" in
   that direction.  However, the viability of the path from station 2 to
   station 6 can only be presumed, unless additional evidence is
   available.  If in fact the header is from an AX.25 I or S frame (but
   not a U frame), an AX.25 virtual circuit has apparently been
   previously established between the end stations and the presumption
   is strengthened.  In this case each link from 4 to 6 is marked
   "synchronized" (but not the link from 2 to 0).

   Not all stations can both originate frames and digipeat them. Station
   4 is observed to originate and station 7 to digipeat, but station 9
   is only a presumptive digipeater and no evidence is available that
   the remaining stations can originate frames.  Thus, the link from
   station 4 to station 7 is marked "source" and from station 7 to
   station 2 is marked "digipeated."

   Depending on the presence of congestion and hidden stations, it may
   happen that the reciprocal path in the direction from station 6 to
   station 4 has quite different link characteristics;  therefore, a
   link can be recognized as heard in each direction independently.  In
   the above diagram the link between 2 and 7 has been heard in both
   directions and is marked "reciprocal".  However, there is only one
   synchronized mark, which can be set in either direction.  If a
   particular link is not marked either heard or synchronized, any
   presumption on its viability to carry traffic is highly speculative
   (the traffic is probably a beacon or "CQ").  If later marked
   synchronized the presumption is strengthened and if later marked
   heard in the reciprocal direction the presumption is confirmed.

   Experience shows that a successful routing algorithm for any
   packet-radio channel must have provisions for congestion avoidance.
   There are two straightforward ways to cope with this.  The first is a
   static measure of node congestion based on the number of links in the
   network graph incident at each node.  This number is computed by the
   wiretap routine and stored in the node table as it adds entries to
   the link table.

   The second, not yet implemented, is a dynamic measure of node
   congestion which tallies the number of link references during the
   most recent time interval (of specified length).  The current plan
   was suggested by the reachability mechanism used in the ARPANET and
   the Exterior Gateway Protocol [3].  An eight-bit shift register for
   each node is shifted in the direction from high-order to low-order
   bits, with zero-bits preceeding the high-order bit, at the rate of
   one shift every ten seconds.  If during the preceeding ten-second

Mills                                                           [Page 6]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

   period a header with a path involving that node is found, the
   high-order bit of the register is set to one.  When a path is
   calculated the number of one-bits in the register is totalled and
   used as a measure of dynamic node congestion. Thus, the time interval
   specified is 80 seconds, which is believed appropriate for the AX.25
   channel dynamics.

5.  Factor Computations and Weights

   The data items produced by the wiretap routine are processed to
   produce a set of factors that can be used by the routing routine to
   develop optimum routes.  In order to insure a stable and reliable
   convergence as the routing algorithm constructs and discards
   candidate paths leading to these routes, the factor computations
   should have the following properties:

   1.  All factors should be positive, monotone functions which increase
       in value as system performance degrades from optimum.

   2.  The criteria used to estimate link factors should be symmetric;
       that is, their values should not depend on the particular
       direction the link is used.

   3.  The criteria used to estimate node factors should not depend on
       the particular links that traffic enters or leaves the node.

   Each factor is associated with a weight assignment which reflects the
   contribution of the factor in the distance calculation, with larger
   weights indicating greater importance.  For comparison with other
   common routing algorithms, as well as for effective control of the
   computational resources required, it may be desirable to impose
   additional restrictions on these computations, which may be a topic
   for further study.  Obviously, the success of this routing algorithm
   depends on cleverly (i.e.  experimentally) determined factor
   computations and weight assignments.

   The particular choices used in the prototype implementation should be
   considered educated first guesses that might be changed, perhaps in
   dramatic ways, in later implementations.  Nevertheless, the operation
   of the algorithm in finding optimum routes over all choices in factor
   computations and weights is unchanged.  Recall that the wiretap
   routine generates data items for each node and link heard and saves
   them in the node and link tables.  These items are processed by the
   routing routine to generate the factors shown below in Table 1 and
   Table 2.

Mills                                                           [Page 7]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

      Factor  Weight  Name            How Determined
      ---------------------------------------------------------------
      f0      30      hop             1 for each link
      f1      50      unverified      1 if not heard either direction
      f2      5       non-reciprocal  1 if not heard both directions
      f3      5       unsynchronized  1 if no I or S frame heard

                         Table 1. Link Factors

      Factor  Weight  Name            How Determined
      ---------------------------------------------------------------
      f4      5       complexity      1 for each incident link
      f5      20      digipeated      1 if station does not digipeat
      f6      -       congestion      (see text)

                         Table 2. Node Factors

   With regard to link factors, the "hop" factor is assigned as one for
   each link and represents the bias found in other routing algorithms
   of this type.  The intent is that the routing mechanism degenerate to
   minimum-hop in the absence of any other information.  The
   "unverified" factor is assigned as one if the heard bit is not set
   (not heard in either direction), while the "non-reciprocal" factor is
   assigned as one if the reciprocal bit is not set (not heard in both
   directions).  The "unsynchronized" factor is assigned as one if the
   synchronized bit is not set (no I or S frames observed in either
   direction).

   With regard to node factors, the "complexity" factor is computed as
   the number of links incident at the node, while the "congestion"
   factor is to be computed as the number of intervals in the eight
   ten-second intervals preceding the time of observation in which a
   frame was transmitted to or through the node.  The "digipeated"
   factor is assigned as one if the node is only a source (i.e.  no
   digipeated frames have been heard from it).  For the purposes of
   path-distance calculations, the node factors are taken as zero for
   the endpoint nodes, since their contribution to any path would be the
   same.

Mills                                                           [Page 8]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

6.  The Routing Routine

   The dynamic data base built by the wiretap routine is used by the
   routing routine to compute routes as required.  Ordinarily, this
   needs to be done only when the first frame to a new destination is
   sent and at intervals thereafter, with the intervals perhaps
   modulated by retry count together with congestion thresholds, etc.
   The technique used is a variation of the Viterbi Algorithm [1], which
   is similar to the the shortest-path-first algorithm used in the
   ARPANET and elsewhere [2].  It operates by constructing a set of
   candidate paths on the network graph from the destination to the
   source in increasing number of hops. Construction continues until all
   the complete paths satisfying a specified condition are found,
   following which one with minimum distance is selected as the primary
   route and the others ranked as alternate routes.

   There are a number of algorithms to determine the mimimum-distance
   path on a graph between two nodes with given metric.  The prototype
   implementation operates using a dynamic path list of entries derived
   from the link table.  Each list entry includes (a) the NID of the
   current node, (b) a pointer to the preceding node on the path and (c)
   the hop count and (d) distance from the node to the final destination
   node of the path:

                   [ NID, pointer, hop, distance ] .

   The algorithm starts with the list containing only the entry [
   dest-NID, 0, 0, 0 ], where dest-NID is the final destination NID, and
   then scans the list starting at this entry.  For each such entry it
   scans the link table for all links with either to-NID or from-NID
   matching NID and for each one found inserts a new entry:

         [ new-NID, new-pointer, hop + 1, distance + weight ] ,

   where the new-NID is the to-NID of the link if its from-NID matches
   the old NID and the from-NID of the link otherwise.  The new-pointer
   is set at the address of the old entry and the weight is computed
   from the factors and weights as described previously.  The algorithm
   coontinues to select succeeding entries and scan the link table until
   no further entries remain to be processed, the allocated list area is
   full or the maximum hop count or distance are exceeded, as explained
   below.

   Note that in the Viterbi Algorithm, which operates in a similar
   manner, when paths merge at a single node, all except one of the
   minimum-distance paths (called survivors) are abandonded.  If only
   one of the minimum-distance paths is required, Wiretap does the same;

Mills                                                           [Page 9]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

   however, in the more general case where alternate paths are required,
   all non-looping paths are potential survivors.  In order to prevent a
   size explosion in the list, as well as to suppress loops, new list
   entries with new-NID matching the NID of an existing entry on the
   path to the final destination NID are suppressed and paths with hop
   counts exceeding (currently) eight or distances exceeding 255 are
   abandoned.

   If the Wiretap station NID is found in the from-NID of an entry
   inserted in the list, a complete path has been found.  The algorithm
   remembers the minimum distance and minimum hop count of the complete
   paths found as it proceeds.  When only one of the minimum-distance
   paths (primary route) is required, then for any list entry where the
   distance exceeds the minimum distance or the hop count exceeds the
   maximum hop count (plus one), the path is abandoned and no further
   processing done for it.  When alternate routes are required the
   hop-count test is used, but the minimum-distance test is not.

   The above pruning mechanisms are designed so that the the algorithm
   always finds all complete paths with the minimum hop count and the
   minimum hop count (plus one), which are designated the alternate
   routes. The assignment of factor computations and weights is intended
   to favor minimum-hop paths under most conditions, but to allow the
   path length to grow by no more than one additional hop under
   conditions of extreme congestion.  Thus, the minimum-distance path
   (primary route) must be found among the alternate paths, usually, but
   not always, one of the minimum-hop paths.

   At the completion of processing the complete paths are ranked first
   by distance, then by the order of the final entry in the list, which
   is in hop-count order by construction, to establish a well-defined
   ordering.  The first of these paths represents the primary route,
   while the remaining represent alternatives should all lower-ranked
   routes fail.

   Some idea of the time and space complexity of the routing routine can
   be determined from the observation that the computations for all
   primary and secondary routes of the example in Appendix A with 58
   nodes and 98 links requires a average of about 30 list entries, but
   occasionally overflows the maximum size, currently 100 entries.  Each
   step requires a scan of all the links and a search (for loops) along
   the maximum path length, which in principle can add most of the links
   to the list for each new hop.  Obviously, the resources required can
   escalate dramatically, unless effective pruning techniques such as
   the above are used.

   The prototype implementation requires 316 milliseconds on an

Mills                                                          [Page 10]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

   LSI-11/73 to calculate the 58 primary routes to all 58 nodes for an
   average of about 5.4 milliseconds per route.  The implementation
   requires 1416 milliseconds to calculate the 201 combined primary and
   alternate routes to all 58 nodes for an average of about 3.4
   milliseconds per route.

7.  Data Base Housekeeping

   In normal operation Wiretap tends to pick up a good deal of errors
   and random junk, since it can happen that a station may call any
   other station using ad-hoc heuristics and often counterproductive
   strategies. The result is that Wiretap may add speculative and
   erroneous links to the data base.  In practice, this happens
   reasonably often as operators manually try various paths to stations
   that may be shut down, busy or blocked by congestion.  Nevertheless,
   since Wiretap operates entirely by passive monitoring, speculative
   links may represent the principal means for discovery of new paths.

   The number of nodes and links, speculative or not, can grow without
   limit as the Wiretap station continues to monitor the channel.  As
   the size of the node table or link table approaches the maximum, a
   garbage-collection procedure is automatically invoked.  The procedure
   used in the prototype implementation was suggested by virtual-memory
   storage-management techniques in which the oldest unreferenced page
   is replaced when a new page frame is required.  Every link table
   entry includes an age field, which is incremented once each minute if
   its value is less than 60, once each hour otherwise and reset to zero
   when the link is found in a monitor header.  When new space is
   required in the link table, the link with the largest product of age
   and distance, as determined by the factor computations and weights,
   is removed first.

   Every node table entry includes the congestion factor mentioned
   above, which is a count of the number of links (plus one) incident at
   that node.  As links are removed from the link table, these counts
   are decremented.  If the count for some node decrements to one, that
   node is removed.  Thus, if new space is required in the node table,
   links are removed as described above until the required space is
   reclaimed.

   In addition to the above, and in order to avoid capture of the tables
   by occasional speculative spasms on one hand and stagnation due to
   excessively stale information on the other, if the age counter
   exceeds a predetermined threshold, currently fifteen minutes for a
   speculative link and 24 hours for other links, the link is removed

Mills                                                          [Page 11]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

   from the data base regardless of distance.  It is expected that these
   procedures will be improved as experience with the implementation
   matures.

8.  Summary and Directions for Further Development

   Wiretap represents an initial experiment and evaluation of the
   effectiveness of passive monitoring in the management of the AX.25
   packet-radio channel.  While the results of initial experiments have
   been encouraging, considerable work needs to be done in the
   optimization effectively, some experience needs to be gained in the
   day-to-day operation of the prototype system during which various
   combinations of weight assignments can be tried.

   The prototype implementation has been in use for about four months at
   this writing;  however, a number of lessons were quickly learned. The
   implementation includes a finite-state automaton to manage initial
   connection requests, including the capability to retry SABM frames
   along alternate routes computed by Wiretap.  A simple but effective
   heuristic is used to generate speculative paths by artificially
   adding links between the destination station and the Wiretap station
   together with all other stations in the node table identified as
   digipeaters.  The algorithm then operates as described above to
   generate the primary and alternate routes.  An example of this
   technique is given in the Appendix.

   This technique works very well, at least in the initial-connection
   phase of virtual-circuit mode, although it requires significant
   computational resources, due to the large number of possible paths
   ranging from reasonable to outrageous.  In the case of datagram mode
   only the primary route is computed.  The heuristic path-abandonment
   strategy outlined above is a critical performance determinant in this
   area.

   While there is a mechanism for the TAPR-1 processor to notify the
   prototype implementation that a lower-level AX.25 virtual circuit has
   failed, so that an alternate path can be tried, there is no intrinsic
   mechanism to signal the failure of an upper-level TCP connection,
   which uses IP datagrams wrapped in AX.25 I frames (connection mode)
   or UI frames (connectionless mode).  This is a generic problem with
   any end-system protocol where the peers are located physically
   distant from the link-level entities.  Experience indicates the value
   of providing a two-way conduit to share control information between
   protocol layers may be seriously underestimated.

   The prototype implementation manages processor and storage demands in
   relatively simple ways, which can result in considerable

Mills                                                          [Page 12]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

   inefficiencies.  It is apparent that in any widely distributed
   version of Wiretap these demands will have to be carefully managed.
   As suggested above, effective provisions to purge old information,
   especially speculative links, are vital, as well as provisions to
   control the intervals between route computations, for instance as a
   function of link state and traffic mode.

   The next step in the evolution towards a fully distributed routing
   algorithm is the introduction of active probing techniques.  This
   should considerably improve the capability to discover new paths, as
   well as to fine-tune existing ones.  It should be possible to
   implement an active probing mechanism while maintaining compatibility
   with the passive-only Wiretap, as well as maintaining compatibilty
   with other stations using no routing algorithms at all.  It does seem
   that judicious use of beacons to discover and renew paths in the
   absence of traffic will be required, as well as some kind of
   echo/reply mechanism similar to the ICMP Echo/Reply support required
   of Internet hosts.

   In order to take advantage of the flexibility provided by routing
   algorithms like Wiretap, it will be necessary to revise the AX.25
   specification to include "loose" source routing in addition to the
   present "strict" source routing.  Strict source routing requires
   every forwarding stage (callsign) to be explicitly declared, while
   loose source routing would allow some or all stages to be left to the
   discretion of the local routing agent or digipeater.  One suggestion
   would be to devise a special collective indicator or callsign that
   could signal a Wiretap digipeater to insert the computed route string
   following its callsign in the AX.25 frame header.

   A particularly difficult area for any routing algorithm is in its
   detection and reponse to congestion.  Some hints on how the existing
   Wiretap mechanism can be improved are indicated in this document.
   Additional work, especially with respect to the hidden-station
   problem, is necessary.  Perhaps the most useful feature of all would
   be a link-quality indication derived from the radio, modem or
   frame-level procedures (checksum failures).  Conceivably, this
   information could be included in beacon messages broadcast
   occasionally by the digipeaters.

   It is quite likely that the most effective application of routing
   algorithms in general will be at the local-area digipeater sites.
   One reason for this is that these stations may have off-channel
   trunking facilities that connect different areas and may exchange
   wide-area routing information via these facilities.  The routing
   information collected by the local-area Wiretap stations could then
   be exchanged directly with the wide-area sites.

Mills                                                          [Page 13]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

9.  References

   [1]  Forney, G.D., Jr.  The Viterbi Algorithm.  Proc IEEE 61, 3
        (March 1973), 268-278.

   [2]  McQuillan, J., I.  Richer and E.  Rosen.  An overview of the new
        routing algorithm for the ARPANET.  Proc.  ACM/IEEE Sixth Data
        Comm. Symp., November 1979.

   [3]  Mills, D.L.  Exterior Gateway Protocol Formal Specification.
        DARPA Network Working Group Report RFC-904, M/A-COM Linkabit,
        April 1984.

   [4]  Fox, T.L., (Ed.).  AX.25 amateur packet-radio link-layer
        protocol, Version 2.0.  American Radio Relay League, October
        1984.

Mills                                                          [Page 14]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

Appendix A.  An Example

   An example will illustrate how Wiretap constructs primary and
   alternate routes given candidate node and link tables.  The candidate
   tables resulted from a scenario monitoring normal traffic on the
   145.01-MHz AX.25 packet-radio channel in the Washington, DC, area
   during a typical 24-hour period.  The node and link tables
   illustrated below give an idea of what the constructed data base
   looks like, as well as provide the basis for the example.

   Figure 1 illustrates a candidate node table showing the node ID
   (NID), callsign and related information for each station.  The Route
   field contains the primary route (minimum-distance path), as a string
   of NIDs from the origination station (NID = 0) to the destination
   station shown, with the exception of the endpoint NIDs.  The absence
   of a route string indicates the station is directly reachable without
   the assistance of a digipeater.  Note that the originating station is
   always the first entry in the node table, in this case W3HCF, and is
   initialized with defaults before the algorithm is started.

      NID Callsign    Flags   Links   Last Rec    Wgt   Route
      -------------------------------------------------------
      0    W3HCF      005     26      15:00:19    255
      1    WB4APR-5   017     18      16:10:38    30
      2    DPTRID     000     3       00:00:00    210   1
      3    W9BVD      005     3       23:24:33    40
      4    W3IWI      015     5       16:15:30    35
      5    WB4JFI-5   017     34      16:15:30    35
      6    W3TMZ      015     2       01:00:49    150   1
      7    WB4APR-6   017     14      14:56:06    35
      8    WB4FQR-4   017     4       06:35:15    40
      9    WD9ARW     015     3       14:56:04    115   11

      10   WA4TSC     015     3       15:08:53    115   11
      11   WA4TSC-1   017     9       15:49:15    35
      12   KJ3E       015     4       15:57:26    155   1
      13   WB2RVX     017     3       09:19:46    135   7
      14   AK3P       015     2       12:57:53    185   7 15
      15   AK3P-5     016     4       12:57:53    135   7
      16   KC2TN      017     3       04:01:17    135   7
      17   WA4ZAJ     015     2       21:41:24    240   5
      18   KB3DE      015     3       23:38:16    35
      19   K4CG       015     3       13:29:14    35

      20   WB2MNF     015     2       04:01:17    180   7 16
      21   K4NGC      015     3       14:57:44    90    8
      22   K3SLV      005     2       03:40:01    160   1

Mills                                                          [Page 15]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

      23   KA4USE-1   017     6       14:57:44    35
      24   K4AF       005     3       12:46:38    40
      25   WB4UNB     015     2       06:45:09    240   5
      26   PK64       005     3       02:50:54    40
      27   N4JOG-2    015     3       13:24:53    35
      28   KX3C       015     4       02:57:29    35
      29   W3CSG      015     4       06:10:17    115   11

      30   WD4SKQ     015     3       16:00:33    35
      31   WA7DPK     015     3       01:28:11    35
      32   N4JGQ      015     3       22:57:50    35
      33   K3AEE      005     3       03:52:43    40
      34   WB3ANQ     015     3       04:01:27    140   7
      35   K2VPR      015     2       12:07:51    240   5
      36   G4MZF      015     3       01:38:30    35
      37   KA3ERW     015     2       03:11:17    155   1
      38   WB3ILO     015     2       02:10:34    140   7
      39   KB3FN-5    016     4       06:10:17    110   11

      40   KS3Q       015     5       15:54:57    35
      41   WA3WUL     015     2       03:36:18    135   7
      42   N3EGE      015     3       15:58:01    160   1
      43   N4JMQ      015     2       08:02:58    185   7 13
      44   K3JYD-5    016     5       15:58:01    155   1
      45   KA4TMB     015     3       16:15:23    115   11
      46   KC3Y       015     2       04:14:36    155   1
      47   W4CTT      005     2       12:21:33    245   5

      52   K3JYD      015     2       02:16:52    155   1
      54   WA5WTF     015     2       02:01:20    240   5
      55   KA4USE     005     3       23:56:02    105   23
      56   N3BRQ      005     2       02:00:36    40
      57   KC4B       015     2       22:10:37    240   5
      58   WA5ZAI     005     2       12:44:03    40
      59   K4UW       005     2       02:36:05    40
      60   K3RH       015     2       01:20:47    135   7
      61   N4KRR      015     3       10:56:50    35
      62   K4XY       015     2       04:53:16    240   5
      64   WA6YBT     015     2       05:13:07    190   7 15

                     Figure 1. Candidate Node Table

   In the above table the Dist field shows the total distance of the
   primary route, the Links field shows the complexity factor, which is
   the number of links incident at that node (plus one), and the Last
   Rec field shows the time (UT) the station was last heard, directly or
   indirectly. The Flags field shows, among other things, which stations

Mills                                                          [Page 16]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

   have originated frames and which have digipeated them.  The bits in
   this field, which is in octal format, are interpeted as follows (bit
   0 is the rightmost bit):

                Bit     Function                       
                --------------------                   
                0       originating station            
                1       digipeater station             
                2       station heard (Last Rec column)
                3       station synchronized connection

   Among the 58 stations shown in Figure 1 are eleven digipeaters, all
   but three of which also originate traffic.  All but twelve stations
   have either originated or digipeated a synchronized connection and
   only one "station" DPTRID, actually a beacon, has not been heard to
   either originate or digipeat traffic.

   Figure 2 illustrates a candidate node table of 98 links showing the
   from-NID, to-NID, Flags and Age information for each link as
   collected. The bits in the Flags field, which is in octal format, are
   interpeted as follows (bit 0 is the rightmost bit):

                          Bit     Function    
                          ------------------- 
                          0       source      
                          1       digipeated  
                          2       heard       
                          3       synchronized
                          4       reciprocal  

      From    To      Flags   Age            From    To      Flags   Age
      ---------------------------            ---------------------------
      5       0       017     0               1       0       037     5
      4       0       015     0               5       4       035     0
      4       1       015     28              7       0       017     60
      9       5       015     60              1       5       006     56
      4       7       015     60              11      0       017     24
      7       15      036     62              7       13      037     60
      12      1       015     71              15      14      035     62
      7       16      037     70              12      5       015     71
      19      0       015     61              16      20      035     70
      5       11      036     60              23      0       017     60
      5       24      035     73              30      0       015     71
      29      11      015     69              5       29      035     73
      8       21      035     67              8       5       017     67
      31      0       015     72              31      5       015     72
      32      0       015     74              32      5       015     69

Mills                                                          [Page 17]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

      40      5       015     17              40      0       015     19
      34      7       015     70              35      5       015     62
      1       40      035     74              38      7       015     71
      5       36      035     72              45      5       015     0
      36      0       015     72              5       30      035     14
      37      1       015     70              44      5       016     14
      12      44      015     17              46      1       015     69
      34      1       015     72              44      1       016     70
      5       23      036     60              9       11      015     79
      10      11      015     60              1       6       035     72
      27      5       015     61              11      1       006     83
      45      11      015     76              52      1       015     71

      5       2       000     14              8       0       005     76
      57      5       015     75              17      5       015     75
      3       0       005     74              3       5       005     74
      26      5       005     71              26      0       005     74
      18      5       015     74              18      0       015     74
      55      5       005     73              24      0       005     62
      61      0       015     63              55      23      005     73
      54      5       015     71              61      5       015     63
      59      0       005     71              56      0       005     71
      5       7       006     71              7       60      035     72
      28      0       015     71              62      5       015     69
      1       7       036     70              28      5       015     71
      7       41      035     70              28      1       015     71
      58      0       005     62              1       22      005     70
      33      7       005     70              33      0       005     70
      64      15      015     69              25      5       015     67
      39      10      035     68              11      39      036     68
      43      13      015     65              29      39      015     68
      40      7       015     62              47      5       005     62
      19      23      015     61              27      0       015     61
      42      1       005     23              23      21      035     60
      1       2       000     5               42      44      015     14

                     Figure 2. Candidate Link Table

   The following tables illustrate the operation of the routing
   algorithm in several typical scenarios.  Each line in the table
   represents the step where an entry is extracted from the path list
   and new entries are determined.  The "Step" column indexes each step,
   while the "To" column indicates the NID of the station at that step.
   The "Ptr" column is the index of the preceeding step along the path
   to the destination, while the "Hop" and "Dist" columns represent the
   total hop count and computed distance along that path.

Mills                                                          [Page 18]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

   Following is a fairly typical example where the destination station
   is not directly reachable, but several multiple-hop paths exist via
   various digipeaters.  The algorithm finds four digipeaters:  1, 5, 11
   and 39, all but the last of which are directly reachable from the
   originating station, to generate two routes of two hops and two of
   three hops, as shown below.  Note that only the steps leading to
   complete paths are shown.

      Destination: 29  Station: W3CSG
      Step    NID     Ptr     Hop     Dist    Comments
      -------------------------------------------------------------
      0       29      0       0       0
      1       5       0       1       30
      2       11      0       1       35
      3       39      0       1       35
      4       0       1       2       235     Complete path: 0 5 29
      35      0       2       2       115     Complete path: 0 11 29
      37      9       2       2       115
      38      10      2       2       115
      39      1       2       2       120
      40      45      2       2       115
      41      39      2       2       110
      42      11      3       2       85
      43      10      3       2       85
      46      0       39      3       240     Complete path: 0 1 11 29
      63      0       42      3       165     Complete path: 0 11 39 29

   The algorithm ranks these routes first by distance and then by order
   in the list, so that the two-hop route at N = 35 would be chosen
   first, followed by the three-hop route at N = 63, the two-hop route
   at N = 4 and, finally the three-hop route at N = 46.  The reason why
   the second choice is a three-hop route and the third a two-hop route
   is because of the extreme congestion at the digipeater station 5,
   which has 34 incident links.

   Following is an example showing how the path-pruning mechanisms
   operate to limit the scope of exploration to those paths most likely
   to lead to useful routes.  The algorithm finds one two-hop route and
   four three-hop routes.  In this example the complete list is shown,
   including all the steps which are abandond for the reasons given.

Mills                                                          [Page 19]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

      Destination: 13  Station: WB2RVX
      Step    NID     Ptr     Hop     Dist    Comments
      -------------------------------------------------------------
      0       13      0       0       0
      1       7       0       1       30
      2       43      0       1       35      No path
      3       0       1       2       135     Complete path: 0 7 13
      4       4       1       2       135
      5       15      1       2       130
      6       16      1       2       130
      7       34      1       2       135
      8       38      1       2       135     No path
      9       60      1       2       130     No path

      10      5       1       2       140     Max distance 310
      11      1       1       2       130
      12      41      1       2       130     No path
      13      33      1       2       140
      14      40      1       2       135
      15      5       4       3       210     Max distance 380
      16      0       4       3       215     Complete path: 0 4 7 13
      17      1       4       3       215     Max distance 305
      18      14      5       3       180     Max hops 4
      19      64      5       3       185     Max hops 4

      20      20      6       3       175     Max hops 4
      21      1       7       3       205     Max distance 295
      22      0       11      3       250     Complete path: 0 1 7 13
      23      4       11      3       255     Max distance 300
      24      12      11      3       255     Max distance 295
      25      40      11      3       250     Max distance 295
      26      37      11      3       255     Max distance 285
      27      46      11      3       255     Max distance 285
      28      44      11      3       255     Max distance 280
      29      34      11      3       255     Max distance 290

      30      6       11      3       250     Max distance 280
      31      52      11      3       255     Max distance 285
      32      28      11      3       255     Max distance 295
      33      0       13      3       215     Complete path: 0 33 7 13
      34      0       14      3       215     Complete path: 0 40 7 13
      35      5       14      3       215     Max distance 385
      36      1       14      3       210     Max distance 300

   The steps labelled "No path" are abandonded because no links could be
   found satisfying the constraints:  (a) to-NID or from-NID matching
   the NID of the step, (b) loop-free or (c) total path distance less

Mills                                                          [Page 20]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

   than 256.  The steps labelled "Max distance" are abandonded because
   the total distance, computed as the sum of the Dist value plus the
   weighted node factors, would exceed 256 as shown.  The steps labelled
   "Max hops" are abandonded because the total hop count would exceed
   the minimum hop count (plus one) as shown.

   Although this example shows the computations for all alternate
   routes, if only the primary route is required all steps with total
   distance greater than the minimum-distance (135) can be abandonded.
   In this particular case path exploration terminates after only 14
   steps.

   The following example shows a typical scenario involving a previously
   unknown station;  that is, one not already in the data base. Although
   not strictly part of the algorithm itself, the strategy in the
   present system is to generate speculative paths consisting of an
   imputed direct link between the originating station and the
   destination station, together with imputed direct links between each
   digipeater in the data base and the destination station.  The new
   links created will time out according to the cache-management
   mechanism in about fifteen minutes.

   In the following example the destination station is 74, which results
   in the following additions to the link table:

      fm-NID  To-NID  Flags   Node Type
      ----------------------------------
      0       74      000     Originator
      1       74      000     Digipeater
      5       74      000     Digipeater
      7       74      000     Digipeater
      8       74      000     Digipeater
      11      74      000     Digipeater
      13      74      000     Digipeater
      15      74      000     Digipeater
      16      74      000     Digipeater
      23      74      000     Digipeater
      39      74      000     Digipeater
      44      74      000     Digipeater

   There are eleven digipeaters involved, not all of which may be used.
   The resulting primary route and five alternate routes are shown
   below.  Note that only five of the eleven digipeaters are used.  The
   remainder were either too far away or too heavily congested.  Note
   that only the list entries leading to complete paths are shown.

Mills                                                          [Page 21]


RFC 981                                                       March 1986
An Experimental Multiple-Path Routing Algorithm

      Destination: 74  Station: CQ
      Step    NID     Ptr     Hop     Dist    Comments
      -------------------------------------------------------------
      0       74      0       0       0
      1       0       0       1       90      Complete path: 0 74
      2       1       0       1       90
      4       7       0       1       90
      5       8       0       1       90
      6       11      0       1       90
      7       13      0       1       90
      8       15      0       1       90
      9       16      0       1       90
      10      23      0       1       90
      11      39      0       1       90
      12      44      0       1       90
      13      0       2       2       210     Complete path: 0 1 74
      29      0       4       2       195     Complete path: 0 7 74
      44      0       5       2       150     Complete path: 0 8 74
      45      0       6       2       170     Complete path: 0 11 74
      60      0       10      2       155     Complete path: 0 23 74

Mills                                                          [Page 22]