Network Working Group T. Imielinski
Request for Comments: 2009 J. Navas
Category: EXPerimental Rutgers University
November 1996

GPS-Based Addressing and Routing

Status of this Memo

This memo defines an Experimental Protocol for the Internet
community. This memo does not specify an Internet standard of any
kind. Discussion and suggestions for improvement are requested.
Distribution of this memo is unlimited.

IANA Note:

This document describes a possible experiment with geographic
addresses. It uses several specific IP addresses and domain names in
the discussion as concrete examples to aid in understanding the
concepts. Please note that these addresses and names are not
registered, assigned, allocated, or delegated to the use suggested
here.

Table of Contents

1. IntrodUCtion...................................... 2
1b. General Architecture...................... 3
1c. Scenarios of Usage: Interface Issues...... 3
2. Addressing Model.................................. 4
2a. Using GPS for Destination Addresses....... 5
3. Routing........................................... 7
3a. GPS Multicast Routing Scheme (GPSM)...... 7
3a-i. Multicast Trees................... 8
3a-ii. Determining the GPS Multicast
Addressing........................ 10
3a-iii. Building Multicast Trees.......... 11
3a-iv. Routing........................... 12
3a-v. DNS Issues........................ 12
3a-vi. Estimations....................... 12
3b. "Last Mile" Routing..................... 13
3b-i. Application Level Filtering....... 13
3b-ii. Multicast Filtering............... 13
3b-iii. Computers on Fixed Networks....... 14
3c. Geometric Routing Scheme (GEO)........... 14
3c-i. Routing Overview.................. 14
3c-ii. Supporting Long-Duration GPScasts. 16
3c-iii. Discovering A Router's Service Area 17

3c-iv. Hierarchical Router Structure and
Multicast Groups.................. 18
3c-v. Routing Optimizations............. 19
3c-vi. Router-Failure Recovery Scheme.... 19
3c-vii. Domain Name Service Issues........ 20
4. Router Daemon and Host Library.................... 21
4a. GPS Address Library - SendToGPS()......... 21
4b. Establishing A Default GPS Router......... 22
4c. GPSRouteD................................. 22
4c-i. Configuration...................... 23
4d. Multicast Address Resolution Protocol (MARP) 23
4e. Internet GPS Management Protocol (IGPSMP). 24
5. Working Without GPS Information................... 25
5a. Users Without GPS Modules................. 25
5b. Buildings block GPS radio frequencies
What then?................................ 25
6. Application Layer Solution........................ 25
7. Reliability....................................... 26
8. Security Considerations........................... 27
9. References........................................ 27
10. Authors' Addresses................................ 27

1. Introduction

In the near future GPS will be widely used allowing a broad variety
of location dependent services such as direction giving, navigation,
etc. In this document we propose a family of protocols and addressing
methods to integrate GPS into the Internet Protocol to enable the
creation of location dependent services such as:

o Multicasting selectively only to specific geographical
regions defined by latitude and longitude. For example,
sending an emergency message to everyone who is currently
in a specific area, such as a building or train station.

o Providing a given service only to clients who are within a
certain geographic range from the server (which may be mobile
itself), say within 2 miles.

o Advertising a given service in a range restricted way, say,
within 2 miles from the server,

o Providing contiguous information services for mobile users
when information depends on the user's location. In
particular providing location dependent book-marks, which
provides the user with any important information which
happens to be local (within a certain range) possibly
including other mobile servers.

The solutions which we present are flexible (scalable) in terms of
the target accuracy of the GPS. We also discuss cases when GPS cannot
be used (like inside buildings).

The main challenge is to integrate the concept of physical location
into the current design of the Internet which relies on logical
addressing. We see the following general families of solutions:

a) Unicast IP routing extended to deal with GPS addresses

b) GPS-Multicast solution

c) Application Layer Solution using extended DNS

The first two solutions are presented in this memo. We only sketch
the third solution.

1b. General Architecture

We will assume a general cellular architecture with base stations
called Mobile Support Stations (MSS). We will consider a wide variety
of cells, including outdoor and indoor cells. We will discuss both
cases when the mobile client has a GPS card on his machine and cases
when the GPS card does not work (i.e. - inside buildings).

We will assume that each MSS covers a cell with a well defined range
specified as a polygon of spatial coordinates and that the MSS is
aware of its own range.

1c. Scenarios of Usage and Interface Issues

Below, we list some possible scenarios of usage for the geographic
messaging.

Consider an example situation, of an area of land near a river.
During a severe rain storm, the local authorities may wish to send a
flood warning to all people living within a hundred meters of the
river.

For the interface to such messaging system we propose to use a zoom-
able map similar to the U.S. Census Bureau's Tiger Map Service. This
map would allow a user to view a geographical area at varying degrees
of magnitude. He could then use a pointing device, such as a mouse,
to draw a bounding polygon around the area which will receive the
message to be sent. The computer would then translate the drawn
polygon into GPS coordinates and use those coordinates when sending
and routing the message. Geographical regions specified using this
zoom-able map could be stored and recalled at a later time. This
zoom-able map is analogous to the IP address books found in many
email programs.

To continue with the above example, local officials would call up a
map containing the river in danger of overflowing. They would then
hand-draw a bounding polygon around all of the areas at least a
hundred yards from the river. They would specify this to be the
destination for a flood warning email to all residents in the area.
The warning email would then be sent. Similar applications include
traffic management (for example, reaching vehicles which are stuck in
traffic) and security enforcement.

Other applications involve general client server applications where
servers are selected on the basis of the geographic distance. For
example, one may be interested in finding out all car dealers within
2 miles from his/her location. This leads to an extension of the Web
concept in which location and distance play important roles in
selecting information. We are currently in the process of
implementing location dependent book-marks (hot lists) in which pages
associated with static and mobile servers which are present within a
certain distance from the client are displayed on the client's
terminal.

2. Addressing Model

Two-dimensional GPS positioning offers latitude and longitude
information as a four dimensional vector:

<Direction, hours, minutes, seconds>

where Direction is one of the four basic values: N, S, W, E; hours
ranges from 0 to 180 (for latitude) and 0 to 90 for longitude, and,
finally, minutes and seconds range from 0 to 60.

Thus <W, 122, 56, 89> is an example of longitude and <N, 85, 66, 43>
is an example of latitude.

Four bytes of addressing space (one byte for each of the four
dimensions) are necessary to store latitude and four bytes are also
sufficient to store longitude. Thus eight bytes total are necessary
to address the whole surface of earth with precision down to 0.1
mile! Notice that if we desired precision down to 0.001 mile (1.8
meters) then we would need just five bytes for each component, or ten
bytes together for the full address (as military versions provide).

The future version of IP (IP v6) will certainly have a sufficient
number of bits in its addressing space to provide an address for even
smaller GPS addressable units. In this proposal, however, we assume
the current version of IP (IP v4) and we make sure that we manage the
addressing space more economically than that. We will call the
smallest GPS addressable unit a GPS-square.

2a. Using GPS for Destination Addresses

A destination GPS address would be represented by one of the
following:

o Some closed polygon such as:

circle( center point, radius )

polygon( point1, point2, point3, ... , pointn)

where each point would be expressed using GPS-square
addresses. This notation would send a message to anyone
within the specified geographical area defined by the closed
polygon.

o site-name as a geographic Access path

This notation would simulate the postal mail service. In
this manner, a message can be sent to a specific site by
specifying its location in terms of real-world names
such as the name of a specific site, city, township,
county, state, etc. This format would make use of the
Directory service detailed later.

For example, if we were to send a message to city hall in Fresno,
California, we could send it by specifying either a bounding polygon
or the mail address. If we specify a bounding polygon, then we could
specify the GPS limits of the city hall as a series of connected
lines that form a closed polygon surrounding it. Since we have a
list of connected lines, we just have to record the endpoints of the
lines. Therefore the address of the city hall in Fresno could look
like:

polygon([N 45 58 23, W 34 56 12], [N 23 45 56, W 12 23 34], ... )

Alternatively, since city hall in Fresno is a well-defined
geographical area, it would be simpler to merely name the
destination. This would be done by specifying "postal-like" address
such as city_hall.Fresno.California.USA.

For "ad hoc" specified areas such as, say a quad between 5th and 6th
Avenue and 43 and 46 street in New York, the polygon addressing will
be used.

Unfortunately, we will not be able to assume that we have enough
addressing space available in the IP packet addressing space to
address all GPS squares. Instead we will propose a solution which is
flexible in terms of the smallest GPS addressable units which we call
atoms. In our solution, a smaller available addressing space (in the
IP packet) will translate into bigger atoms. Obviously, we can use
as precise addressing as we want to in the body of the geographic
messages - the space limitations apply only to the IP addressing
space.

By a geographic address we mean an IP address assigned to a
geographic area or point of interest. Our solution will be flexible
in terms of the geographic addressing space.

Below, we will use the following two terms:

o Atoms: for smallest geographic areas which have
geographic address.

Thus, atoms could be as small as GPS squares but could be
larger

o Partitions: These are larger, geographical areas, which will
also have a geographic address. A state, county, town etc.
may constitute a partition. A partition will contain a number
of atoms.

Here are some examples of possible atoms and partitions:

o A rectangle, defined by truncating either longitude or
latitude part of the GPS address by skipping one or more
least significant digits

o A circle, centered in a specific GPS address with a
prespecified radius.

o Irregular shapes such as administrative domains: states,
counties, townships, boroughs, cities etc

Partitions and Atoms (which are of course special atomic partitions)
will therefore have geographic addresses which will be used by
routers. Areas of smaller size than atoms, or of "irregular shape"
will not have corresponding geographic addresses and will have to
handled with the help of application layer.

3. Routing

Let us now describe the suggested routing schemes responsible for
delivering a message to any geographical destination.

We will distinguish between two legs of the connection from the
sender to the receiver: the first leg from the sender to the MSS
(base station) and the second leg from the MSS to the receiver
residing in its cell. Our two solutions will differ on the first leg
of the connection and use the same options for the second leg, which
we call "last mile".

3a. GPS-Multicast Routing Scheme

Here, we discuss the first leg of routing: from the sender to the
MSS. We start with the multicasting solution.

Each partition and atom is mapped to a multicast address. The exact
form of this mapping is discussed further in this subsection. We
first sketch the basic idea.

This solution provides flexible mix of the multicast and application
level filtering for the geographic addressing. The key idea here is
to approximate the addressing polygon of the smallest partition which
contains it and using the multicast address corresponding to that
partition as the IP address of that message. The original polygon is
a part of the packet's body and the exact matching is done on the
application layer in the second leg of the route.

How is the multicast routing performed?

3a-i. Multicast Trees

The basic idea for the first level of routing using multicast is to
have each base station join multicast groups for all partitions which
intersect its range. Thus, MSS is not only aware of its own range
but also has a complete information about system defined partitions
which its range intersects. This information can be oBTained upon MSS
installation, from the geographic database stored as a part of DNS.

If the proper multicast trees are constructed (using for example link
state multicast protocol) than the sender can simply determine the
multicast address of the partition which covers the original polygon
he wants to send his message to, use this multicast address as the
address on the packet and put the original polygon specification into
the packet content. In this way, multicast will assure that the
packet will be delivered to the proper MSS.

Example

For instance the MSS in New Brunswick may have its range intersect
the following atoms and partitions: Busch, College Avenue, Douglass
and Livingston Campuses of Rutgers University (atoms), New Brunswick
downtown area (atom), the Middlesex county partition and the NJ state
partition. Each of these atoms and partitions will be mapped into a
multicast address and the New Brunswick's MSS will have to join all
such multicast groups.

The message will be then specified and sent as follows:

The user will obtain the map of the New Brunswick area possibly from
the DNS extended properly with relevant maps. He will specify the
intended destination by drawing a polygon on the map which will be
translated into the sequence of coordinates. In the same time the
polygon will be "approximated" by the smallest partition which
contains that polygon. The multicast address corresponding to that
partition will be the IP address for packets carrying our message.
The exact destination polygon will be a part of each packet's body.
In this way the packet will be delivered using multicast routing to

the set of MSS which are members of the specified multicast group
(that is all MSS whose ranges intersect the given partition). Each
such MSS now will follow the "last mile" routing which is described
in detail, further in the proposal. Briefly speaking, the MSS could
then multicast the message further on the same multicast address and
the client will perform the final filtering o application layer,
matching its location (obtained from GPS) with the polygon specified
in the packet's body. Other solutions based entirely on multicasting
are also possible as described below.

End_Example

However, things cannot be as simple as described. For such a large
potential number of multicast groups if we build entire multicast
trees, the routing tables could be too large. Fortunately it is not
necessary to build complete multicast trees. Indeed, it in not
important to know precise location of each atom in California, from a
remote location, say in NJ.

Thus, we modify our simple solution by implementing the following
intuition:

The smaller is the size of the partition (atom) the more locally is
the information about that partition (atom) propagated.

Thus, only multicast group membership for very large partitions will
be propagated across the whole country.

For example, a base station in Menlo Park, California can intersect
several atoms ) and several larger which cover Menlo Park, such say
a partition which covers the entire San Mateo county, next which
cover the entire California and finally next which may cover the
entire west coast. This base station will have to join multicast
groups which correspond to all these rectangles. However, only the
information about multicast group corresponding to the West Coast
partition will be propagated to the East Coast routers.

However, a simple address aggregation scheme in which only a "more
significant portion" of address propagates far away would not work.
Indeed, in this case a remote router, say in NJ, could have several
aggregate links leading to California - in fact, in the worst case,
all its links could point to California since it could have received
a routing information to some location in California on any of those
links.

To avoid this, for each partition we distinguish one or a few MSS
which act as designated router(s) for that partition. For example,
the California partition, may have only three designated routers, one

in Eureka, another in Sacramento and yet another in LA. Only the
routing entries from the designated routers would be aggregated into
the aggregate address for California. Information coming from other
city routers will simply be dropped and not aggregated at all. This,
in addition to a standard selection of the shortest routes, would
restrict the number of links which lead to an aggregate address. In
particular, when there is only one designated router per partition,
there would only be one aggregate link in any router. This could lead
to non-optimal routing but will solve the problem of redundant links.

Even with a designated routers, it may happen that the same packet
will arrive at a given base station more than once due to different
alternative routes. Thus, a proper mechanism for discarding redundant
copies of the same packet should still be in place. In fact, due to
the possible intersections between ranges of the base stations the
possibility of receiving redundant copies of the same packets always
exist and has to be dealt with as a part of any solution.

3a-ii. Determining the geographic Multicast Addressing

Here we describe more specifically, the proposed addressing scheme
and the corresponding routing.

The addressing will be hierarchical. We will use the following
convention - each multicast address corresponding to a partitions or
an atoms will have the following format:

1111.GPS.S.C.x

where GPS is the specific code corresponding to the geographic
addressing subspace of the overall multicast addressing space. The S,
C and x parts are described below:

S - Encoding of the state.
Each state partition will have the address S/0/0.

C - County within a state.
Each county partition having the address S/C/0.

x - Atom within a county.

where 0's refer to the sequences of 0 bits on positions corresponding
to the "C part" and "x part" of address.

For example if GPS part is 6 bit,s which gives 1/64 of existing
multicast addresses to the geographic addressing we have 22 bits
left. The S part will take first 6 bits, C part next 6 bits (say)
and then the next 10 bits encode different atoms (within a county).

Thus, in our terminology the proposed addressing scheme has two types
of partitions: states and counties.

We will assume that the GPS network will consist of all base stations
(MSS) in addition the rest of the fixed network infrastructure. The
designated GPS routers however, will only be selected from the
population of MSS. Specifically, there will be state dedicated and
county dedicated routers.

The concept of the designation will be implemented as follows. From
the set of all MSS, only certain MSS will play a role of designated
routers for county and state partitions. Non-designated MSS will
only join multicast groups which correspond to the GPS atoms but not
GPS partitions that they intersect. The MSS which is a designated
router for a county partition will join the multicast group of the
county in which it is located, but not the state. Finally the state
designated router will also join the multicast address corresponding
to the state it is located in.

3a-iii. Building Multicast Trees

We assume that each router has geographic information attached to it
- in the same format as we use for multicast mapping, S/C/x - it
encodes the atom that contains the router.

The multicast tree is built by a router propagating its multicast
memberships to the neighboring routers. A given router will only
retain certain addresses though, to follow the intuition of not
retaining a specific information which is far away.

This is done as follows: the router (not necessarily the MSS based
router) with the address S/C/x will only retain addresses about
S'/0/0, S/C'/0 for S' and C' different from S and C and S/C/x for all
x. Thus, it will drop all the addresses of the form S'/C'/y for all
S' different that S except those with C'=0 and y=0, as well as all
the addresses of the form S/C'/y with C' different from C except
those with y=0. Hence, these addresses will not be forwarded any
further either.

Thus, notice that only the information coming from designated routers
will be forwarded further away, since the non-designated routers are
not allowed to join the multicast groups which correspond to the
states and counties. Consequently, their multicast membership
information will be not be propagated.

In this way a router at S/C/x will not bother about specific
locations within S'/C'/y since they are "too far".

Notice that this service may not be provided everywhere so we may not
have to use all multicast addresses even within those assigned for
geographic addresses.

Notice also that all of this is flexible - if we have more multicast
addresses available (IP v 6) we will get more precise addressing due
to smaller atoms.

3a-iv. GPS Routing

Given a packet we always look for the "closest" match in the routing
table. If there is a complete match we follow such a link, if not we
follow the address with the x-part 0'd in (county address) if there
is none with the county which agrees with the destination county than
we look at the entry which agrees with the state part of the
destination address.

3a-v. DNS Issues

How does the client find out the multicast address on which the
packet is to be sent? We assume that the local name server has the
complete state/county hierarchy and that each county map can be
provided possibly with the "grid" of atoms and partitions already
clearly marked.

Points of interests within a county can be attached multicast address
just as atoms. Then a given base station would have to join multicast
groups of the points of interests that it covers.

The final stage is for the receiver to look at the polygon (point of
interest) which is encoded in the body of the multicast packet and
decide on the basis of its own GPS location if this packet is to be
received or not. Doing it on the application layer simplifies many
routing issues. There is a tradeoff, however, specially when we have
very short S/C/x addresses and base stations which do not cover the
given polygon in fact are reached unnecessarily. This may happen and
it needs to be determined what is the number of the multicast
addresses which are necessary to reduce this "false" alarms to the
minimum.

3a-vi. Estimations

Assume average cell size of, say, 2km x 2km and the average state
size: say 200,000 square km, the average county size: say 4,000
square km.

A reasonable size of the atom is around the size of the cell since
then we do not hit wrong cells too often.

Therefore we need the x addressing part of the S/C/x to encode
4,000/4 cells: 1.000 atoms. Thus we need 10 bits for x part. With 6
bits for the state and 6 bits for the county that gives 22 bits which
is 1/64 of the total IP v4 multicast addressing space.

With IPv6 we will have, of course, much more addressing space which
we can use for the GPS multicast routing.

3b. "Last Mile" Routing

Multicasting will be used for the last mile routing in both our
solutions (i.e the one just discussed and the geometric routing
solution described next), but in different ways.

3b-i. Application Level Filtering

The MSS will forward the geographic message on its wireless link
under a multicast address. This multicast address will either be the
same for all locations in the range of the MSS's cell or, there will
be several addresses corresponding to atoms which intersect the given
cell. Additionally, a complete GPS address (for example in the form
of the polygon) will be provided in the body of the packet and the
exact address matching will be performed on the application layer.
The receiver, knowing its GPS position uses it to match against the
polygon address. The GPS position can be obtained by the receiver
either from the GPS card or, indoors, from the indoor base station
which itself knows its GPS position as a part of configuration file.

3b-ii. Multicast Filtering

In multicast level filtering, the base station assigns a temporary
multicast address to the addressing polygon in a message. It will
send out a directive on the cell's specially assigned multicast
address. All mobile clients who reside in that cell are members of
that special multicast group (one per MSS). The directive sent by the
MSS will contain the pair consisting of the temporary multicast
address together with the polygon. To improve the reliability this
message will be multicast several times. The clients, knowing their
GPS positions will than join the temporary multicast groups if their
current locations are within the advertised polygon. The MSS will
then send out the real message using the temporary multicast address.

The temporary multicast address would be cached for a period of time.
If more packets for the same polygon arrive in a short period of
time, they will be sent out on the same multicast address. If not,
then the multicast address is dropped and purged from the cache.
Filtering on the client's station is then performed entirely on the
IP level. This solution introduces additional delay (needed to join

the temporary multicast group) but reduces the number of irrelevant
packets received by the client. This especially important for very
long messages.

3b-iii. Computers on Fixed Networks

Fixed-network computers should also monitor all of the mandatory
multicast addresses for their site and GPS square. In this manner,
the fixed computers will also receive messages sent to specific GPS-
addresses.

Modified base stations would still be in charge of multicasting the
messages to the computers. These base stations would have the same
GPS-routing functionality as the mobile computer base stations.
Their main difference would be that the mobile computer base stations
would use radio frequencies to multicast their messages and the fixed
network base stations use the local Ethernet or Token Ring network.

The next scheme differs from the GPS multicast scheme described above
only on the first leg of the route, from the sender to the MSS. The
"last mile" from the MSS to the final destination will have the same
options as described above.

3c. Geometric Routing Scheme (GEO)

The Geometric Routing Scheme (GEO) uses the polygonal geographic
destination information in the GPScast header directly for routing.
GEO routing is going to be implemented in the Internet Protocol (IP)
Network layer in a manner similar to the way multicast routing was
first implemented. That is, a virtual network which uses GPS
addresses for routing will be overlayed onto the current IP
internetwork. We would accomplish this by creating our own GPS-
address routers. These routers would use tunnels to ship data
packets between them and between the routers and base stations.

3c-i. Routing Overview

Sending a GPScast message involves three steps: sending the message,
shuttling the message between routers, and receiving the message.

Sending a GPScast message is very similar to sending a UDP datagram.
The programmer would use the GPScast library routine SendToGPS().
Among other parameters, this routine will accept the GPS polygonal
destination address and the body of the message. The SendToGPS()
routine will encapsulate the GPScast message in a UDP datagram and
send it to the class E address 240.0.0.0. Previously, the system
administrator will have specified in the /etc/rc.local or /etc/rc.ip
file a route command that will specify that packets with the address

240.0.0.0 will instead be sent to the address of the local GPS
router. This will have the effect of sending the datagram to the
nearest GPS router.

Before explaining how the GPS routers shuttle the GPScast message to
its destination, an introduction to routers and their different parts
is in order. For scalability purposes, GPS routers are arranged in a
hierarchical fashion. Each layer would correspond to a distinct
geographic area, such as a state or a city. At the top would be
country-wide routers in charge of moving messages from one end of the
country to another. At the bottom would be campus or department
routers in charge of moving messages between the base stations. See
Figure 1.

Country-Router(s)
/
State-Router(s)
/
City-Router(s)
/
Router Router
/
Base Base Base Base Base

Figure 1: Hierarchy of routers.

A GPS router essentially consists of three parts: a service area
table containing the geographic area serviced by the router and each
of its hierarchical children, a hashed cache of previous actions, and
a table containing the IP addresses of at least the router's children
and the router's parent. In the case of a bottom-layer campus
router, the service area table will contain polygons describing the
geographic reach of each child base station's cell. The polygon
created from the union of all of the router's child base stations'
polygons defines the service area of the router.

Once the datagram arrives at a GPS router, the router strips the
datagram off, thereby, leaving it with the original GPScast message.
First the router must determine if it services any part of the area
of the destination polygon. To do this, the router finds the
intersection between the destination polygon and the polygon
describing the router's service area. The polygon intersection
algorithm used is described by O'Rourke in his paper, A New Linear
Algorithm for Intersecting Convex Polygons. This algorithm requires
order N-squared time in the worst case. If the intersection result
is null, then the router simply sends the message to its parent
router.

------ Destination Polygon
A
--------------
B Router's Service Area Polygon
--------------
C
------

Figure 2: Polygon Difference

However, if the result is not null, then the router does service the
area described by the intersection polygon. The router now subtracts
its service area from the destination polygon and sends the rest to
it's parent router. This subtraction step is actually a by-product
of the intersection algorithm. Using the example in Figure 2, the
destination polygon and the router's service area polygon intersect
at the region labeled B. Therefore, the router will subtract out the
B section and send the remaining sections A and C to its parent
router.

Continuing with the example, the router now uses the intersection
polygon B to to determine which base station (or stations) will
receive the GPScast message. The router finds the intersection
between the region B and the polygon of each base station's cell.
Those base station polygons which intersect the region B will be sent
the GPScast message. Processes on Mobile Hosts serviced by these
base stations will now use the routine RecvFromGPS() to receive the
GPScast message.

3c-ii. Supporting Long-Duration GPScasts

Most likely, there will be a need to support sending real-time
continuous media to a GPS destination. This continuous media could
be an audio GPScast or a video GPScast. This would require that
jitter be reduced in order to minimize disturbing artifacts in the
audio or video playback. Continually checking the destination
geometry of each packet would incur unnecessary delays and may
promote jitter.

Therefore, the router will keep a hashed cache of the latest GPScast
packets and their destinations. Each cache item will be hashed using
the Sender Identification included in the header of GPScast messages
as the key. Each cache item will contain a time stamp and a list of
the next hops for that GPScast. When the time stamp exceeds a
certain limit, then the cache item will be dropped. The list of next
hops is a list of the IP addresses of the base stations, peer
routers, and parent router which are to receive a copy of the GPScast
messages.

When a router receives a GPScast packet, it will use the incoming
packet's Sender Id as a key into the hashed cache. If this is not
the first packet to arrive for this destination and if the timer on
the hash table entry has not yet expired, then the hashed cache will
return a list of all of the destination addresses to which copies of
the packet must be sent. Copies of the packet are sent to all of
these destinations and the hash entry's time stamp is updated.

If no hash table entry is found (i.e.- this is the first packet
encountered for this destination address), then the normal geometry
checking routine would take over. A new cache entry is made
recording all of the next-hop destination addresses of the GPScast.
In this manner, if several other packets with the same GPS
destination follow this first packet, the router can use the hash
table to look-up the destination base stations instead of calculating
it using geometry.

3c-iii. Discovering A Router's Service Area

When the router is initiated, it will consult its configuration file.
One of the items it will find in the file will be the multicast
address of the base station group to which all of its child base
stations are members. The router will join this group and then send
out Service Area Query messages to this multicast group periodically
to discover and to refresh its knowledge of its children base
stations and the geographical areas serviced by them.

Queries are issued infrequently (no more than once every five
minutes) so as to keep the IGPSMP overhead on the network very low.
However, since the query is issued using unreliable multicast
datagrams, there is a chance that some base stations may not receive
the query. This is important in two cases: when a child node fails
and when a router first boots up. The case of a failed child node
will be explained later. However, when a router first boots up, it
can issue several queries in a small amount of time in order to
guarantee that base stations will receive the query and to,
therefore, build up its knowledge about its child base stations
quickly.

Base stations respond to a Service Area Query by issuing a Service
Area Report. This report is issued on the same multicast group
address that all of the base stations have joined. The report
contains the geographical service area of the base station. In order
to avoid a sudden congestion of reports being sent at the same time,
each base station will initiate a random delay timer. Only when the
timer expires will the base station send its report.

For every base station that responds, the router will create an IP
tunnel between it and the base station. This tunnel will carry the
GPScast packet traffic between the base station and the router. Each
responding base station and its geographic area of service will also
be included in the router's geometric routing table as a possible
destination for GPScast packets. Any base station that does not
respond for ten continuous Service Area Queries will be considered
unreachable and will be dropped from the routing table.

3c-iv. Hierarchical Router Structure and Multicast Groups

R5----------------------R6
/ /
R1---------R2 R3---------R4
/ / / /
b1 b2 b3 b4 b5 b6 b7 b8 b9 b10 b11 b12

Figure 3: Two peer routers (R5 and R6) cooperatively servicing four
child routers (R1 - R4).

For scalability purposes, a hierarchy of routers is used to transport
messages from a sender to a receiver. Each layer of peer routers
would have its own multicast group address for the exchange of
Service Area Queries and Reports between the peer routers. However,
routers in distinct subtrees need not know about the routers in other
subtrees. Therefore, multicast group addresses will also differ
between hierarchy subtrees. See figure 3. For instance, routers R1
and R2 would share a multicast group and would know about each other.
At the same time, routers R3 and R4 would share a different multicast
group and would know about each other. However, routers R1 and R2
would not know about R3 and R4, and vice versa.

But how will the router know the location and number of its peer
routers and who its parent router is? As mentioned before, the
router consults its configuration file upon start-up. Included in
this configuration file will be the the address of its parent router
and the multicast group address that the peer routers will use. This
peer multicast group address will be used in the same manner as the
base station multicast group address. It will be used to send and
receive Service Area Queries and Reports between the parent router
and the peer routers. There is only one difference. When a router
sends a Service Area Report, in addition to reporting its
geographical service area, a router will include the multicast
address of its children base stations. The reason for this is
explained in the router-failure recovery scheme described below.

3c-v. Routing Optimizations

The optimization described here attempts to reduce the latency of a
GPScast. It does so by reducing the the number of hops a packet must
traverse before finding its destination. The intuition behind the
idea is this: instead of going to the parent router and then to the
sibling, simply go to the sibling directly. As an additional
benefit, this method prevents the parent router from becoming a
bottleneck or a point of failure in the routing scheme.

In this optimization, when a router attempts to determine who will
receive the GPS packet, it considers its peer routers as if they were
also its children in the routing hierarchy. This means that the
router will consider its service area to be the union of the service
areas of its children and its peer routers. Also, when the
destination polygon intersects the router's service area polygon, the
router will forward a copy of the GPScast packet to any child or peer
router whose geographic service area contains or touches the packet's
GPS destination polygon.

However, before it sends a copy of the packet to a peer router, it
first finds the polygon:

P = D / S

where D stands for the packet's destination GPS polygon, S is the
polygon representing the service area of the peer router, and P is
the polygon that represents the intersection of D and S. The polygon
P is substituted for the destination polygon D in the packet and only
then is the packet forwarded to the peer router. This is necessary
because the peer router will be using that same routing algorithm.
Therefore, if the peer router receives a packet with the original
destination polygon D, it will also route copies of the packet to all
of its qualifying peer routers causing a chain of packet copies being
bounced back and forth.

3c-vi. Router-Failure Recovery Scheme

In the case of a router failure, the system should be able to route
around the failed router and continue to service GPScast messages.
The responsibility of detecting whether a router has failed or not
falls to the parent router. Using Figure 3 as an example router
hierarchy, the parent router R5 periodically sends out Service Area
Query IGPSMP messages on its children's multicast group address.
Thus, the child routers R1 and R2 will both receive this query.
Normally, both routers will respond with a Service Area Report
message. This message contains a polygon describing their service
areas and the multicast group address of their children.

However, if a router, R1, does not respond to ten continuous queries,
then it must be considered to have failed. Upon detecting this, the
parent router R5 will send a Set Service Area message to the child
router, R2 telling it to assume responsibility for the base stations
underneath the failed R1 router. In this Set Service Area message,
the parent router includes the multicast group address of R1's
children. The R2 router uses this multicast address to learn the
service areas and IP addresses of R1's children. The R2 router then
issues a Service Area Report advertising its new enlarged service
area responsibilities. All peer and parent routers will then update
their routing tables to include this new information. When the
failed router, R1, restarts, it will declare that it is alive and
that it is again servicing its area. All routers will then again
update their routing tables.

In the case that there is no parent router, such as at the top of the
routing hierarchy, then each peer router will keep track of its
neighbors. If a neighbor router fails, then the first neighbor
router to declare that it is taking over the base stations for the
failed router will take responsibility. The rest continues as
before.

3c-vii. Domain Name Service Issues

Domain Name Servers (DNS) could be used to facilitate the use of GPS
geographic addressing for sites of interest. The aim is to describe
specific geographic sites in a more natural and real-world manner
using a postal-service like addressing method. Essentially, the DNS
would resolve a postal-service like address, such as
City_Hall.New_York_City.New_York, into the IP address of the GPS
router responsible for that site. The GPS router would then route
the message to all available recipients in the site.

The DNS would be used when a message is sent using the

site-code.city-code.state-code.country-code

addressing scheme. The DNS would evaluate the address in reverse
starting with the country code, then the state code, etc. This is
the same method used currently by the IP DNS service to return IP
addresses based on the country or geographic domains.

4. Router Daemon and Host Library

4a. GPS Address Library - SendToGPS()

A library for GPS address routing will be constructed. The main
routines contained in this library will be the SendToGPS() and
RecvFromGPS() commands. SendToGPS() has the following syntax:

SendToGPS(int socket, GPS-Address *address, char *message, int size)

where socket is a previously created datagram socket, address is a
filled GPS-Address structure with the following form:
typedef _GPS-Address {
enum { point, circle, polygon } type;
char *mail-address;
struct
{
enum { North, South, West, East } dir;
int hours, minutes, seconds;
} *points;

} GPS-Address;

and message and size specify the actual message and its size. The
SendToGPS() routine will take the GPS-addressed message, encapsulate
it in an IP packet, and then send it as a normal IP datagram. The
message is encapsulated in the following manner:

--------------------------------------------------------
IP Header with destination address set to 240.0.0.0
--------------------------------------------------------
Sender Identifier
--------------------------------------------------------
Address Type - CirclePolygon
--------------------------------------------------------
Actual GPS Address (see below)
--------------------------------------------------------
Body of Message
--------------------------------------------------------

where the Sender Identifier would consist of a combination of the
sender's process id, host IP address, and the center of the
destination polygon. The Actual Address would be one of the
following:

circle - single GPS address and range measured in centiminutes.

polygon - list of GPS addresses terminated by the impossible
address: N 255 255 255.

RecvFromGPS() has the following syntax:

RecvFromGPS(int socket,GPS-Address *address,char *message,int size)

where socket is a previously created datagram socket, address is an
empty GPS-Address structure, and message and size specify message
buffer and its size.

4b. Establishing A Default GPS Router

The default GPS router is determined using the unicast routing table
found in the UNIX kernel. The local system administrator will have
previously adjusted the table so that all GPScast messages are sent
to the local GPS router. However, if there is no route for GPScast
messages in the table, then all messages will, by default, be sent to
the default gateway. If the default gateway does not support GPScast
messages, then all attempts to send a GPScast will return an error.

By default, all GPScast messages will initially have as their
destination the class E address 240.0.0.0. A route will be added to
the kernel routing table by the system administrator for this
address. The route will specify the location of the local GPS
router. The "route" command will be used to affect the routing table
and it can be placed in the /etc/rc.local or /etc/rc.ip files so that
it will take effect each time the computer is booted. For example,
to specify that GPScast messages addressed to 240.0.0.0 should, by
default, be sent to the router which resides on a computer on the
same subnet with local address 128.6.5.53, use the following:

/etc/route add host 240.0.0.0 128.6.5.53 0

If the default destination for GPScast messages is a host that does
not support GPS addressing, then Network Unreachable errors will be
returned to any process attempting to route GPScasts through that
host.

4c. GPSRouteD

In order to provide the capability of GPS address routing throughout
an IPv4-based internetwork, special-purpose routers will be created
to support GPS address routing on top of the current Internet. These
routers, which will be called GPSRouteD, will use virtual point-to-
point links called tunnels in order to connect two GPSRouteDs
together over regular unicast networks. The tunnels work by
encapsulating the GPS address messages in IP datagrams and then

transmitting the message to the host on the other end of the tunnel.
In this manner, the GPS address messages look like normal unicast
packets to all IPv4 routers in between the two GPS address routers.
At the end of the tunnel, the receiving GPSRouteD removes the GPS
address message from the datagram and continues the routing process.

By using tunnels, the GPS routers can be established as a virtual
internetwork throughout the current Internet without regard for the
physical properties of the underlying networks. Moreover, the use of
tunnels means that the host on which the router daemon is running
need not be connected to more than one subnet in order for the router
to forward GPS messages. This virtual internetwork would be
responsible for routing GPS address messages only. This virtual
network, however, is not intended to be a permanent solution and is
only intended to provide a means of supporting GPS address routing
until it gains wider acceptance and support in the Internet
infrastructure.

4c-i. Configuration

When a GPSRouteD initially executes, it first checks the file
/etc/GPSRouteD.conf for configuration commands to add tunnel and
multicast links to other GPS address routers. There are two kinds of
configuration commands:

multicast <multicast-address> <peerchild>

tunnel <local-addr> <remote-addr>
<parentpeerchildhost> <service-area>

The tunnel command is used to create a tunnel between the local host
on which the GPSRouteD executes and a remote host on which another
GPSRouteD executes. The tunnel must be set up in the GPSRouteD.conf
files at both ends before it will be used.

The multicast command tells the router which multicast addresses to
join. These addresses will carry IGPSMP messages and replies. The
router will use these IGPSMP messages to build up and keep current
its own internal routing table.

4d. Multicast Address Resolution Protocol (MARP)

Of course, this begs the question, how will the individual computers
know which multicast addresses to join? For example, an MH would
have to join the multicast address of its current cell so that it can
receive GPScast messages (using application-level filtering) or
directions to join other multicast groups (using multicast
filtering). We have designed a protocol called Multicast Address

Resolution Protocol (MARP) that works the same way as Reverse Address
Resolution Protocol (RARP). However, instead of returning the IP
address of the MH, it will return multicast group address of the cell
the MH is currently in. The MH would then join this multicast group.

4e. Internet GPS Management Protocol (IGPSMP)

The Internet GPS Management Protocol (IGPSMP) is used by GPS routers
to report, query, and inform their router counterparts about their
geographical service areas. The IGPSMP will also be used to verify
that routers are correctly functioning.

The vocabulary of IGPSMP will consist of six Words:

o set service area - Used by the parent router to set the
geographic service area of a router. This is needed in
order to automatically respond to router failure or new
router boot-up.

o confirm service area - confirms that a router has received
its service area.

o geographical service area query - This message will be used
by a router to build up its geographical routing table.
It is sent to all routers on the same level.

o service area report - This message is sent in response to a
query request. It contains a bounding closed polygon
described using GPS coordinates which contains the service
area for the router.

o ping - This message is sent periodically to ascertain whether
the router is currently functioning properly. Usually sent
by the parent router in the hierarchy tree.

o alive signal - Usually sent as a reply to the ping message.
Used by a router to indicate that it is functioning
correctly. It is also sent immediately after a router
boots.

All of IGPSMP messages will be sent on an all-routers multicast
address for a particular hierarchy level. The exact multicast
address can be set in the router configuration file.

Note that for the GPS-Multicast routing scheme, the time-to-live
value of the service area reports will be varied in order to control
the distribution of the information. In GPS-Multicast routing, only

the multicast group membership for very large partitions will be
distributed throughout the country. Smaller partition may only be
distributed to neighbor routers.

5. Working Without GPS Information

5a. Users Without GPS Modules

Mobile users without GPS modules can still participate - though at a
very reduced level. When an MH enters a cell, it can use an MARP to
discover the local multicast group for that cell or atom. As the
user roams from cell to cell, the mobile host can keep track of the
current cell that the user is in and adds or drops the multicast
groups pertaining to those cells. The user's GPS address can be set
to be the center of the current cell.

5b. Buildings block GPS radio frequencies. What then?

Each room can have a radio beacon placed on the ceiling. The beacon
will be weak enough so that it will not penetrate walls. Each radio
beacon will have its own GPS-address associated with it which it will
broadcast. When a mobile user enters a room, his MH will detect the
beacon and read the beacon's GPS address. The GPS-address of the MH
will be set to the GPS-address of the beacon. The MH will then use
this beacon's GPS address in order to perform any message filtering
that it needs to do. Now the mobile user can have a GPS-address
associated with him even though he is indoors and his GPS-module is
useless.

6. Application Layer Solution

In this subsection we sketch a third solution which relies more
heavily on the DNS.

In the application layer solution the geographic information is added
to the DNS which provides the full directory information down to the
level of the IP address of each base station and its area of coverage
represented as a polygon of coordinates.

A new first level domain - "geographic" is added to the set of first
level domains. The second level domain names include states, the
third, counties and finally, the fourth: polygons of coordinates, or
so called points of interests. We can also allow, polygons to occur
as elements of second, third domains to enable sending messages to
larger areas.

Thus a typical geographic address can look like

city-hall-Palo-Alto.San-Mateo-County.California.geographic

or

Polygon.San-Mateo-County.California.geographic

where Polygon is a sequence of coordinates.

This geographic address is resolved in a similar way as the standard
domain addresses are resolved today into a set of IP addresses of
base stations which cover that geographic area. There are several
possibilities here:

a. A set of unicast messages is sent to all base stations
corresponding to the IP addresses returned by the DNS. Each base
station then forwards the message using either of the two last link
solutions: application level or network level filtering.

b. All the base stations join the temporary multicast group for the
geographic area specified in the message. In this way we may avoid
sending the same message across the same link several times. Thus,
after the set of relevant base stations is determined by the DNS, the
temporary multicast group is established and all packets with that
multicast address are sent on that multicast address.

c. Only one, central to the polygon base station is returned by the
DNS just as in the IP unicast solution. However that "central" base
station will have to forward messages to the other base stations
within the polygon.

Notice that we should distinguish between "small area" and "wide
area" geographic mail. The "small area" mail will be most common and
will most likely involve just one base station, favoring a simple
form of solution (a).

7. Reliability

Should the geographic messages be acknowledged?

Since we have no control if users are present in the target
geographic area where the mail is distributed we do not see a need
for individual acknowledgments from the message recipients. However,
we believe that the base stations (MSS) covering the target area of
geographic mail should acknowledge the messages.

Typically only a few base stations will be involved since typically
we will not cover very broad geographic areas anyway. We assume that
the base stations, additionally to forwarding the the messages on
their wireless interfaces will buffer them, either to periodically
multicast them (emergency response) or to provide them to users who
just entered a cell and download the "emergency stack" of messages
for that area as a part of the service hand-off protocol.

8. Security Considerations

Some method of determining who has permission to send messages to a
large geographical area is needed. For instance, perhaps only the
mayor of New York City has permission to send a message to all of New
York City.

9. References

Deering, S., "Host Extensions for IP Multicasting", STD 5, RFC1112,
August 1989.

S. Deering. Multicast Routing in a Datagram Internetwork. Ph.D.
Thesis, Stanford University, (December 1991).

J. O'Rourke, C.B. Chien, T. Olson, and D. Naddor, A new linear
algorithm for intersecting convex polygons, Computer Graphics and
Image Processing 19, 384-391 (1982).

J. Ioannidis, D. Duchamp, and G. Q. Maquire. IP-Based Protocols for
Mobile Internetworking. Proc. of ACM SIGCOMM Symposium on
Communication, Architectures and Protocols, pages 235-245,
(September, 1991).

10. Authors' Addresses

Tomasz Imielinski and Julio C. Navas
Computer Science Department
Busch Campus
Rutgers, The State University
Piscataway, NJ
08855

Phone: 908-445-3551
EMail: {imielins,navas}@cs.rutgers.edu