Beruflich Dokumente
Kultur Dokumente
J a c o b Sharony
Hnzcltine C o r p o r a t i o n
G r e e n l a w n , NY 11740
sharony@hazeltinc.com
0-7803-3250-4/96$5.00O1996 IEEE
807
ity of physical subnet 1 2 (i.e., within “hearing” distance from
its members) it will keep its current address. Any node in the
network is updated with the current addresses used in its phys-
ical and virtual subnets (by its logical neighbors). This can be
accomplished, e.g., by an advertising process where each node
notifies its logical neighbors of its current address using a ded-
icated management channel. Therefore, a node which desires
to join a specific physical subnet would contact a member(s) of
this physical subnet t o find out which address it can acquire,
and then would advertise its newly acquired address t o all of
its logical neighbors. Note t h a t if a node cannot reach any of
its logical neighbors in its virtual subnet,, it will use another
virtual subnet via one of its logical neighbors in its physical
subnet. This case will be further discussed in Section 5 dealing
with routing.
Observe t h a t there is no logical connection between the vir-
tual subnets, however, since they are “overlaid” on t h e same
Figure 1: Physical and virtual subnets in a mobile radio net- region they might interfere with each other. Thik interference
work. is eliminated by the channel access scheme in use. One way,
for example, is t o operate each virtual subnet on a different fre-
quency channel(s); when the number of frequency channels is
possible is N = p q . Each node in this topology is affiliated less than the number of virtual subnets some form of time shar-
with nodes whose address differs only in one digit; t h a t is, node ing can be used. Note t h a t a less acute problem exists between
~ 1 . is
~ 0affiliated with nodes Z I . Z ~0, 5 X; 5 q - l , z ; # 2 0 , neighboring physical subnets which have a limited degree of
and with nodes 3c;.zo, 0 5 xi 5 p - 1,xi # 2 1 . T h u s , every overlapping since each one of them covers a limited area. Thus,
node is affiliated with p + q - 2 other nodes; we say t h a t each one can take advantage of spatial reuse, where only neighboring
+
node has p q - 2 logical neighbors. Next we group every q (overlapping) subnets use different frequency channels.
nodes t h a t differ only in their LSD into an MSD group, and
every p nodes t h a t differ only in their MSD into an LSD group.
+
Note t h a t there are altogether p q groups, and each node is a 4 Mobility management
member of one LSD group and one MSD group. These groups A mobile node which changed its subnet affiliation will notify
are the basic building blocks of the network as described in the
all the nodes in its new physical and virtual subnets (i.e., its
next section. current logical neighbors) of its newly acquired address. This
notification process can take place, e.g., during the establish-
3 Logical topology ment of links with its logical neighbors or by broadcasting in
its physical and virtual subnets. In general, a source node does
Each node in the network is affiliated with a physical subnet not know the current address of a desired destination node.
(MSD group) and a virtual subnet (LSD group). Nodes which T h e source node can determine this address by inquiring in its
are members of a physical subnet are within close proximity in physical (virtual) subnet, since one of the nodes there is affil-
a local geographic area. Nodes which are members of a vir- iated with the destination node virtual (physical) subnet. T o
tual subnet form a regional network (i.e., beyond a local area). clarify, let source node S and destination node D addresses be
Figure 1 depicts a mobile radio network with physical subnets SI.SOand Dl.Do, respectively, and denote by IS1I the cardinal-
(in shaded areas) and virtual subnets (e.g., in solid and dashed ity (number of members) of physical subnet SI.Consider two
lines). Note t h a t all nodes within a physical subnet have the different cases for finding node D’s address; first, if ]Si( 2 Do,
same MSD while all nodes within a virtual subnet have the node S would inquire in its physical subnet SI about node D
same LSD. I t is assumed for the moment t h a t nodes of a given and receives node D’s address from node 5’1 .Do (which was no-
physical subnet can reach (e.g., by adjusting their transmission tified earlier by node D , via virtual subnet Do, regarding its
power or by using a directional antenna) nodes of neighboring current address). Second, if I S11 < D O ,node S would inquire
physical subnets. Later we deal with the case when this as- in virtual subnet SO about node D and receives node D’s ad-
sumption does not hold (i.e., when a node is disconnected from dress from node D1 .SOwhich is affiliated with node D physical
its virtual subnet). subnet. Note t h a t node S does not know a-priori which of the
A node becomes a member of a physical subnet by acquiring above cases is valid, nevertheless, it inquires about node D first
the first available address (with the lowest LSD) in t h a t subnet, in its physical subnet and if it does not get a response it inquires
e.g., if a node joins physical subnet 1 2 and there are already in its virtual subnet ( a t least one of node S logical neighbors
10 members in this subnet it will use the address 12.10 since knows node D’s address). Alternatively (instead of inquiring
LSD’s 0-9 are occupied already. Once a node becomes affili- about node D’s address), node S can broadcast its packets for
ated with a specific physical subnet, automatically it becomes node D in its physical and virtual subnets, a t least one node
a member of a virtual subnet defined by the LSD in its address; (which is a logical neighbor of node D ) will be able t o forward
referring t o t h e above example, the node will be a member in the packets t o their destination.
virtual subnet 10. As long as the node remains in t h e vicin- Figure 2 describes a simplified location updating and track-
808
loc-track I
I
809
L1 total number of liiiks activated in t h e physical subnets
during Phase I.
7' - load of link i in on? large mnlti-hop network, i.e., t,he num-
ber of times link i is traversed by all possible N ( N - 1)
paths in t,he network.
that, each r o u t e traverses alt,ernately physical and virt,iial sill>- 711 maximum link-load during Phase I (i.e.. m a x J ( q i ) ) ,
nets. It can be shown t h a t between each source-dest#inationpair
+
there are p (I - 2 disjoint paths, i.e., paths t h a t do not share 772 ~ masirnuin link-load during Phase I1 (i.e.? maxk(y;))
links or nodes (e.g., for p = (I = fl the number of disjoint
It is assumed t h a t traffic is homogeneous, where each node
paths is O(fl)). Each of these paths corresponds t o one of in the network sends X packets/sec t,o any of t,lie other N - 1
+
the p Q - 2 logical rieighbors of the source riode. Note that, a nodes. 'ro simplify the presentation, w e use an M / M / 1 queue-
pat,h is uniquely specified oncr a logical neighbor was srlected ing model to drscribe t,he behavior of each act,ivat,edlink; there-
by t h e source node. T o route a packet from a source node t o fore, the average delay of a packet traversing link IC is given by
a destination riode, the source node selects (say at random) b k = F-, where l / p is t,he average packet length in bits,
one of tlie p +
q - 2 disjoint, paths. In case of i t path failure,
c k is the link capacity in bitslsec and q k is the link load. Note
t h e source node can select (say raridonily) one of the remaining
that if a more accurat,c model for the link behavior is used it.
disjoint paths.
will only result in a different expression for 612. Using Little's
formula a n d suninling over all the activated links iri the net-
6 Performance work, the average queueing delay across t h r net,work is givcn
by
Network p e r f o r m m m in terms of throughput and fault toler
ance is evaluated. We compare the throughput of the network
t o t h a t of a one large multi-hop network
6.1 Throughput
We assume i,liat i i i each multi-hop subnet ii link activation
T D M A - F D M h (multiple frequency channels per subnet are
possible) access scheme is usecl where each link is activatecl Kote that tlie capacity of the links is inversely proportional
at, lrast~once during each t,irne frame. We furtlier a.ssume t h a t
*
t.0 the riurnber of time slots ( i n t h e corresponding phase) which
the above shortest-path routing is used (i.e.. any path traverses depends on t8henurnbrr of freqiiency channels used. T h e maxi-
a t most, two subnets. one physical and one virtual). The anal- mnm traffic between any t,wo nodes during phase I is X I =
ysis in this siibsection is t,riie foi- any roiit.ing procedurr w t t h s n similarly. the maximum t,raffic twt,wprn any two nodes during
t h e mult,i-hop subnets. In t,he following aualvsis we i i s r t,hrse
notations:
phase I1 is A-, = w. 'I'hrrefore, the maximum t.raffic be-
twren a n y t w o nodes in t,hc network is
R - bit rate of each transmitting node.
cy - portion of time frame usecl in Phase I (for int,ra-sribriet
traffic).
T h e normalized network throughput is giveii by
T - number of time-slots rised in one large multi-hop network.
Y ( N - 1)
TI- number of time-clots usecl in Phase I.
r = X(A - 1)-,,,A, -
~
(3)
ILR Ti11i~+ r l ; ~ 2 ~
'1)L - niimbrr of t,inie-slot,s i i s e d in I'hase II Kote t,hat r / ~ and 712 depend on the logical t,opology and t,he
I, ~ total number of links activated in one large multi-hop net- routing procedure usecl within the subnets.
work.
81 0
Figure 5: 16-node network with a well controlled 1 opology. Figure 6: Links activated in the physical subnets.
81 1
architecture is highly fault-tolerant, has a relatively simple lo-
there are six disjoint paths between cation updating and tracking scheme, and by virtue of its load
nodes 1.3 and 3.2
balancing feature, typically achieves a network with relatively
1 3 ---> 1 0 ---> 3 0 -> 3 2 high throughput and low delay. Network performance in terms
1 3--> 1 1 -->3 1 - > 3 2
of throughput and fault-toleiance was evaluated
13->12->32
1 3 --> 0 3 --->0 2 -> 3 2
1 3 --->2 3 --22 2 -> 3 2
13-->33->32 Refer en ce s
[l] E M. Gafni and D P Bertsekas, “Distributed algorithms
for generating loop-free routes in networks with frequently
changing topology,” IEEE Trans. Commun., vol. C O M 29,
pp. 11-18, 1981.
Figure 9 Example node-disjoint paths in a 16-node mobile [2] I Cidon and M Sidi, “Distributed assignment algorithms
radio network using Long-path routing. for multihop packet radio networks,” I E E E Trans C o m p ,
vol. C-38, no. 10, pp. 1353-1361, 1989.
node c o n n e c t i v i t y - K as t h e minimum number of faulty nodes [3] 0. S.d e Souza, P. Sen, and R. R. Boorstyn, “Congestion
t h a t creates a disconnected network. Assuming one-hop sub- based routing in packet radio networks,” in Proc. I E E E
nets, each node in the logical topology is connected directly t o ICC ’89, vol. 3, (Boston), pp. 51.3.1-51.3.5, June 1989.
p + q - 2 other nodes, thus, t h e node connectivity of the net-
[4] R. L. Hamilton, J r . and H. C. Yu, “Optimal routing in
work is 6 = p + q - 2. For example, consider a 1024-node
multihop packet radio networks,” in Proc. I E E E I n f o c o m
network composed of 32 subnets of 32 nodes each, then any 61
’90, June 1990.
nodes can be faulty before the network becomes disconnected.
Define t h e network link connectivity- CT as t h e minimum num- [5] L. H u , “Distributed code assignments for CDMA packet
ber of node-disjoint paths between any source-destination pair radio networks,” I E E E Trans. Network., vol. 1, Dec. 1993.
(i.e., paths t h a t do not share links or pass through the same
[6] M. S. Corson and A. Ephremides, “A distributed rout-
node). T h e link connectivity of the topology is CT = 2 for path-
ing algorithm for mobile wireless networks,” Wireless Net-
lengths of not more than two hops. If we allow path-lengths t o
works, vol. 1, Jan. 1995.
be u p t o three hops (using Long-path routing, see Figure 4 ) , it
can be shown t h a t t h e link connectivity reaches its maximum [7] L. Kleinrock and S. J., “Optimum transmission radii for
value of = p + q - 2 . Referring to the above example, there packet radio networks or why six is a magic number,” Nut.
are a t least 62 node-disjoint paths, i.e., alternative paths be- Telecom. Conf., Dec. 1978.
tween any source-destination pair with path-lengths of a t most
[8] H. Takagi and L. Kleinrock, “Optimal transmission ranges
three hops. Since the network has high node connectivity and
for randomly distributed packet radio terminals,” I E E E
high link connectivity it is therefore very reliable. A possible
technology t h a t can benefit from this is wireless ATM, where Trans. C o m m u n . , vol. COM-32, Mar. 1984.
a virtual path established between a given source-destination [9] T. Hou and V. 0. Li, “Transmission range control in
pair has many alternative disjoint routes which can be used in multihop packet radio networks,” I E E E Trans. C o m m u n . ,
case of node or link failures due to mobility, interference etc. vol. COM-34, Jan. 1986.
Figure 9 depicts an example of a 16-node network ( p = q = 4)
having six node-disjoint paths between a pair of nodes when [lo] L. H u , “Topology control for multihop packet radio net-
using Long-path routing. works,” I E R E Trans. C o m m u n . , vol. 41, Oct. 1993.
[Ill M. Gerla and J . T. Tsai, “Multicluster, mobile, multimedia
7 Conclusions radio network,” Wireless Networks, vol. 1, O c t . 1995.
[la] I. Chlamtac and A. Lerner, “A link allocation protocol for
An architecture comprising a logical topology of physical and
mobile multihop networks,” in Proc. I E E E Globecoin ’85,
virtual subnets, and corresponding addressing, mobilit,y man-
Dec. 1985.
agement and routing schemes were described. This architecture
is applicable to mobile radio networks and accommodates dy- [13] J . Sharony and A. C. Sevdinoglou, “Distributed TDMA-
namic topology changes due to relative movement of network FDMA-CDMA link assignment in mobile radio networks
nodes. T h e architecture partitions a mobile network into log- with/without flexible directivity.” To be submitted for
ically independent subnetworks. Network nodes are members publication, 1996.
of physical and virtual subnets and may change their affilia-
tion with these subnets due to thrir mobility. Each node is
allocated an address based on its current subnet affiliation. We
observed - especially in large networks with random t,opology -
t h a t partitioning of the network may result in significantly more
balanced load than in one large multi-hop network, an attribute
t h a t can significantly improve the net,work’s performance. T h e
812