Digitized by the Internet Archive in 2013 http://archive.org/details/faulttolerantgri1030lies UIUCDCS-R-80-1030 t3 UILU-ENG 80 1729 FAULT-TOLERANT GRID BROADCASTING by Arthur L. Liestman ^^ August 1980 ***** so UIUCDCS-R-80-1030 FAULT -TOLERANT GRID BROADCASTING by Arthur L. Liestman August 1980 DEPARTMENT OF COMPUTER SCIENCE UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN URBANA, IL 61801 This project is funded in part by NASA Grant NSG 1471. Page 1 1 Introduction * Consider a graph G = (V,E) which represents a communications network. The set V of vertices corresponds to the members of the network and the set E of edges corresponds to communication lines connecting pairs of members. One member of the network has a message which is to be disseminated to all the other members. This is to be accomplished as quickly as possible by a series of calls over the network. The series of calls is constrained by the follow- ing: 1. each call requires one unit of time 2. any member may participate in at most one call per time unit 3. a member can only call an adjacent member This process is referred to as ( local ) broadcasting [Mitchell & Hedetniemi, 78]. In the event that a communications link fails the message may not be disseminated to all of the network members. By incorporating redundancy in the calling scheme the completion of the broadcast may be guaranteed in the pres- ence of up to k faults. Several aspects of the broadcasting problem have been studied. Broad- casting in general graphs has been studied in [Farley, 79], [Mitchell & Hedet- niemi, 78], and [Farley, et al. , 79]. Of particular interest here is broad- casting in grid or grid-like graphs which has been studied in [Farley & Hedetniemi, 78], [Cockayne & Hedetniemi, 78], and [Van Scoy, 79]. Grids are of particular interest due to their structural regularity and frequent use in parallel multiprocessor architectures. Schemes are known for minimum time broadcasting from any node of a simple grid [Farley & Hedetniemi, 78]. Such schemes are also known for two types of wraparound grids [Farley & Hedet- niemi, 78]. The maximum number of vertices of an m-dimensional grid graph which may be informed in n time Intervals Is known [Peck, 80]. Page 2 This paper investigates the problem of fault-tolerant broadcasting in grid or grid-like graphs. Fault-tolerant broadcasting [Liestman, 80] is sim- ply broadcasting with redundancy in order that certain faults, i.e. connec- tions between nodes ceasing to function, do not affect the completion of the broadcast. This concept is defined more formally in Chapter 2. In Chapter 3 fault-tolerant broadcast schemes are given for any node in a simple grid graph. In Chapter 4 fault-tolerant broadcast schemes are given for nodes in wraparound grids. 2 Definitions . 2 . 1 Fault-tolerant Broadcasting . Consider a graph G = (V,E) and a subgraph G' = (V,E') with E' = E-E" where E" is a set of k edges of E. The graph G represents a communications network as before. The set of edges E" represents faulty communication links so that the subgraph G' represents the actual communications network. Broad- casting on G with enough redundancy so that the broadcast is completed when any set E" of k links ceases to function is termed _k fault-tolerant broadcasting . The broadcast scheme does not incorporate fault detection. The standard broadcast problem corresponds to k=0. When describing a broadcast scheme for a graph G, a call between two adjacent vertices u and v at time t is represented by labelling the edge (u,v) with t. If u« is the message originator and u. is another vertex of G, a sequence of edges OtyUj), (Uj.iij), ..., (u^.u^ labelled t p tj t ± respectively with t.2 vertices. The complete rectangular grid graph of size mxn , denoted R , is defined to be the 'product' [cf. Harary, 69] of P and P . For example the graph below is the grid R, ,: ' T T T T ' > 1< ii a ii 1 > < I I II II ' o * i ■ i < Each vertex in R can be identified by a pair of integer coordinates (i,j) where Ki5 is m+n-1 and there is a m,n calling scheme which completes 1 fault-tolerant broadcasting in minimum time. Proof : Without loss of generality the corner vertex is (1,1). Clearly m+n-1 time units are required for 1 fault-tolerant broadcasting from (1,1) since labelling two calling paths (of length m+n-2) from (1,1) to (m,n) according to the broadcasting constraints requires at least m+n-1 time units. The following calling scheme which completes 1 fault-tolerant broadcast from (1,1) in m+n-1 time units completes the proof. vertex time talks to (1,1) 1 (1,2) 2 (2,1) (l,j) j-l (I,j-D Kj4 is n+1 and there is a cal- ling scheme which completes 1 fault-tolerant broadcasting in minimum time. Proof : Without loss of generality the corner vertex is (1,1). Clearly n+1 time units are required since labelling two calling paths from (1,1) to (2,n) according to the broadcasting constraints requires at least n+1 time units. The following calling scheme which completes 1 fault-tolerant broadcast from (1,1) in n+2 time units completes the proof: vertex time talks to (l.D 1 2 (2.1) (1.2) Cl.j) Kj4 is n+3 and there is a cal- j,n ling scheme which completes 1 fault- tolerant broadcasting in minimum time. Proof ; Without loss of generality the corner vertex is (1,1). Assume that the broadcast can be done in time < n+3. Consider two calling paths to (3,n). Any labelling of two such paths requires at least n+2 time units. If the two paths meet at a common vertex other than the endpoints the labelling requires more than n+2 time units. The following are the only possible shapes that two non-intersecting paths can take from (1,1) to (3,n): <> — ► o .J' f — ' For each shape there are two labellings of the edges so that (3,n) is informed by both paths by time n+2. It Is easily checked that in each case the label- ling cannot be extended so that the message can reach (l,n) from (2,1) before n+3. The following calling scheme completes the proof: vertex (1.1) (2,1) time 1 2 2 3 4 talks to (1,2) (2,1) (1,1) (3,1) (2,2) vertex time talks to (l.J) j-l (l.J-1) Kj 9 (,' . 8 3 8 -U > r < Corollary 2.*3_ : The 1 fault-tolerant broadcast scheme above requires 6n-5 calls a-s compared to 3n-l calls in the fault-tolerant broadcast scheme. Theorem 3*4: The minimum number of time units necessary for 1 fault-tolerant broadcasting from a corner vertex of R, for n>6 is n+3 and there is a cal- **, n ling scheme which completes 1 fault-tolerant broadcasting in minimum time. Proof : Without loss of generality the corner vertex is (1,1) time units are required to label calling paths to (4,n). Page 11 Clearly n+3 The following calling scheme completes the proof: vertex time talks to (1. 1) 1 ( 2 ( :i,2) :2,d (2, 1) 2 ( 3 < 4 ( :i,d :3,d :2,2) (3, 1) 3 ( 4 < 5 ( 9 ( :2,d :3,2) :4,d [3,2) (4, 1) 5 ( 6 I 9 < :3,d :4,2) :4,2) (3, 2) 4 1 5 1 6 ( 8 1 9 1 :3,d :3,3) :2,2) :4,2) :3,d (4,2) 6 < :4,d 7 1 :4,3) 8 1 :3,2) 9 I :4,d (2, n-1) n+1 1 n+2 1 '2,n- :2,n) 2) n+3 1 :i,n- 1) (3, ,n-l) n+1 1 '3,n- 2) n+2 [4, n- 1) n+3 i :3,n) (1 >n) n-1 n 1 n+3 :2,n) (2,n) 1) (4 ,n-l) n+2 n+3 (3,n- (4,n) 1) vertex U,j) Kj,j) (4,j) 2 Corollary 3_.4_: The 1 fault-tolerant broadcast scheme above requires 8n-6 calls as compared to An-1 calls in the fault-tolerant scheme. Theorem 3_.jk The minimum number of time units for 1 fault-tolerant broadcast from a corner vertex of R for the cases: m,n 1. m=2, n=2,3 2. m=3, n=3 3. m=4, n=4,5 is m+n and there is a calling scheme which completes 1 fault-tolerant broad- cast in minimum time. Proof : The proof that m+n units are required in each case is omitted. The arguments are similar to those above. The following schemes complete the proof: Page 13 ■i-^2- *,5 t » > a 3 3 ■ ■ v • 6 3 6 . -5 . 3 i ^JL. /" D ' < i 2 j i ■» . Z 3 s 8 8 b 3 6 7 i 5 8 , 5 ■ * , 6 ^.8 . ^ / . 2 . 3 . S a 36 ? V s 9 8 58 6 ? 3 8 £ 6 ( * 8 i 8 9 i 1 7 i « 7 , 6 . ► Corollary 3_.J5: The 1 fault-tolerant schemes above require the following number of total calls: 1 . mn+n- 1 2. 2mn-m-l 3. 2mn-2m 4. 2mn-m-l for m=2, n=2, 3 for m=3, n=3 for m=4, n=4 for m=4, n=5 as compared to mn-1 for the fault-tolerant scheme. 3.2 Fault-tolerant Broadcast from a Side Vertex. The above results can be used to construct schemes for 1 fault-tolerant broadcasting from a side vertex. Without loss of generality the message ori- ginator is vertex (i,l) with Ki<[m/2J. The vertex (i,l) is the corner vertex of two subgrids ^ n and ^.^+1 n which share the edges and vertices in row i. For the remainder of this section we will refer to the former as the upper Page 14 subgrid and the latter as the lower subgrid. The two share row 1 which Is row 1 of each of the subgrids. The general approach for broadcasting from (i,l) will be to use the corner scheme for R ... given above for the lower m-i+l,n ° subgrid and a modified version of the corner scheme for R. given above for l,n the upper subgrid. As an illustration, consider vertex (4,1) of Rg ,. as the message origina- tor. The lower subgrid is R, _. The broadcast in the lower grid is as fol- o, :> lows: B ' . X 3 H 1 p—- ■ ■ < 3 5 1 H- 8 I ? - f S 8 , f 3 5 , 7 8 f f H 5 ■ 7 6 8 9 ? 10 5 (p r f , 8 . , 7 . e 9 5 9 , 6 , h 7 8 , 8 , 1 1 6> ?■ 7- 8 , 3 5 3 3 6 ^ 5 2. 5" r 6 ( 8 , * 4 ? 5 9 3 4 s & 6 H 7 j 6 9 s 8 , 5 m=4, n>6 m>4, n=2 for m=2, 3, n=2 m=3, n>3 m=4, n=»4,5 Page 16 The value is not apparent for other values of m and n since the known schemes for these grids are not of the proper form. Theorem 3^6_: m+n+1 m=2, n=3 t(m,n) = m+n m=2, n>4 m>4, n=3 m>5, n=4 Proof ; Consider such a broadcast from the corner of R« «. In order to get two paths to (1,3) by time 5 the following calls are required: L-3- There is no extension of this labelling which includes 2 calling paths to (2,1) by time 5. The following scheme informs all vertices by time 6: *- / , -a 3> Consider a grid R with n>4. Clearly if the top row is labelled with ° 2,n ' r 1, 2, 3, • .., n-1 then a second path cannot reach (l,n) until time n+2. The following scheme informs all vertices by time n+2: «) 1 f J 5 (. (i 1 «! 3,fr H n-3 f "~*\ I °"' I i i 1 n-l n n-hi Consider a grid R _ with m>4. To label 2 calling paths to (m, 3) requires at m, 5 Page 17 least m+2 units. In order to extend such a labelling so that there are also two calling paths to (m, 1) at least m+3 units are required. The following scheme informs all vertices of R „ by time m+3: m,3 ' / 2 3 5 3 l 5 ^ , b H 4 l* , * , 5 , »»»,*»» , , mtl t»H-l ultZ wtt rr,-l OO mo+i mn*3 Consider a grid R , with m>5. To label two calling paths to (m, 3) requires at least m+3 units. In order to extend such a labelling so that there are also two calling paths to (m, 1) at least m+4 units are required. The follow- ing scheme informs all vertices of R , by time m+4: in } "f „ / 3 3 (r - i 3 t 5 b , ? , 1 < ■ 3 * 5 b . f « , V- S 6 Page 18 m-A «»-l ► " 1+ » l i . "lO r*-l 1 IW1 ' w+i , "H-3 . 4 1 ' r<+3 r ni - t '' 2 - ■ r v^t-3 ^-^ *v> vn-H m+T- Corollary 3^6_: The 1 fault-tolerant schemes given above with the edges of row 1 labelled 1, 2, ..., n-1 require the following number of total calls as com- pared to mn-1 in the fault-tolerant schemes: 3n-l for m>l, n=2 9 m"2, n=3 3n-l m=2, n>3 13 m=3, n=3 6n-5 m=3, n>3 5m-l m>3, n=3 24 nr=4, n=4 35 m=4, n«5 8n-6 m=4, n>5 6m- 1 m>4, n= 3 4 2nm-2m-l m>4, n>4 Theorem 3_.7j There Is a calling scheme which completes 1 fault-tolerant broad- cast from a side vertex (i,l) with Ki<[m/2) of R in minimum time. L J m,n Proof : In any scheme for 1 fault-tolerant broadcasting from (i,l) there must be two calling paths to each of the corners (m,n) and (l,n). To label 2 paths Page 19 to (m,n) requires m-i+n time units as the distance from (i,l) is m-i+n-1. If m=2i-l then the distances to the two corners are equal and at least i+n units are required to label the paths to the corners. Thus any scheme must take m- i+n time units (or i+n if m=«2i-l) to complete 1 fault-tolerant broadcast on R m,n A calling scheme can be constructed for any side vertex of R as ° m,n described above. The method for combining the corner schemes for two subgrids is as follows: 1. apply the 1 fault-tolerant corner broadcast scheme for R ... with row 1 edges labelled 1, 2, . .., n-1 to the m— l+i, n lower subgrid. 2. apply the 1 fault-tolerant corner broadcast scheme for R. with row 1 edges labelled 1, 2, ..., n-1 to the upper subgrid with the following modifications: a. all calls except those between row 1 vertices which occur at time t are delayed until time t+1. b. unless 1=2 and n>3 all calls between (l,j) and (2,j) which occur later than time n+1 (after the delay) are deleted. Using the 1 fault-tolerant corner broadcast schemes with row 1 labelled 1, 2, ..., n-1 which attain the broadcast times t(i,n) and t(m-i+l,n) a scheme for 1 fault-tolerant broadcasting from (1,1) is constructed. It is easily verified that there are two calling paths to each vertex and that none of the broadcast rules are violated. The time required by this scheme is easily seen to be no greater than max{t(m-i+l,n), t(I,n)+l}. The time required by the constructed schemes Is actually: i+n+2 i+n+1 if i=2, m=3,4 n=3 if i=2, m=3, n>3 i>3, m=2i-l, n-3 i>4, m=2i-l, n-4 i=4, m=7,8, n=5 1=2, ra=4, n=2 Page 20 i+n if m- i+n+1 m-i+n if if i=2 , m=3, n=2 1=3 . m=5, n=2 1=3 , m=5, n>4 i=3 , m=6, n>6 1=4, m=7, n>6 i>4 , m=2i- -1, n=2 i>5 , m=2i- -1, n>5 1=2, m>5, n=3 i>2 , m>2i , n=4 i=3 , m=6, n=5 1=3, m>6, n=3 i>4 m>2i , n=3 1=2, m=4, n>4 1=2 , m=5, n=2 1=2, m=5, n>6 1=2, m>6, n>5 1=3, m>6, n=2 1=3, m>7, n>5 1=4, m>9, n=5 i>4 , m>2i > n=2 1=4, m>8, n>6 i>5 , m>2i , n>5 In some cases (such as 1=2, m=3, n=2) the resulting time is less than max{ t(m-i+l,n), t(i,n)+l} due to the deletion of calls by step 2b. The schemes constructed for the cases i=2, m=3,4, n=3 and i=3, m=3, n>3 can all be improved. The constructed schemes are: 1? .3 t 1 ' i 31 i z i . * H 3.5 ^ 5 3> 1 a 4 1 2 J * H**. 5 .6 3 c f i ' « r ^ , ,-3 i 7 ' 7 5 C i V 1 i — i 1 , 1 .< ,s "» , r>+i n + a ■*-3 n+3 . — — i i < n+a 1 1 \rt n-f/ Page 21 In the first two cases the deletion of labels by 2b allows a relabelling of other edges resulting in the following schemes: V. <. 5 . ft 5 «- JAi 3.5 *~ jjhk_l In the case 1=2, m=3, n>3 the following relabelling is possible: <^- 3,fe A^^ ri n+/ n+z r%-C »1 n n-3 ( »-l 1 o-l *+l M+Z o+i *i-l v\ ' »+i ' Thus i+n+1 units are used by the schemes for 1=2, m=3,4, n=2 and i+n units are used for 1=2, m=3, n>3. The schemes which take m-i+n units and the schemes which take I+n units if m=2i-l are minimum time schemes. The other cases can also be shown to be as fast as possible. Consider, for example, the case 1=3, m>6, n=3. The lower subgrld is R, _ with k>3. The proof of Theorem 3.6 shows that to com- plete 2 calling paths to both (k,l) and (k,3) requires k+3 units. Thus k+3 = m-i+n+1 units are needed for the side vertex problem. Similar arguments hold for the remaining cases. Page 22 3 . 3 Fault-tolerant Broadcasting from an Interior Vertex . As In the previous section the corner broadcast schemes may be combined to create a 1 fault-tolerant broadcast scheme for interior vertices of a grid. Without loss of generality the message originator is vertex (i,j) with Ki<[m/2] and Kj<[n/2J. The vertex (i,j) is the corner vertex of four subgrids R. . , R, . , , , R ,., ., and R ... „ . , , . The general approach for & i,j i,n-j+l' m-i+l,j' m-i+l,n-j+l ° rr broadcasting from (i,1) will be to use the corner scheme for R ... ... 6 ,J m-i+l,n-j+l given above and modified versions of the corner schemes for the other subgrids. The corner schemes given above which have the edges of row 1 labelled 1, 2, ..., n-1 also have the edges of column 1 labelled 2, 3, . . . , m. This will be required for the general approach. Let t(ra,n,i,j) be the minimum time required for 1 fault-tolerant broad- casting from an interior vertex (i,j) of R with Ki<[m/2] and Kj<[n/2j. id * n Theorem 3.8: t(m,n,i,j) > m+n-i-j+1. Proof : The distance from (i,j) to (m,n) is m+n-i-j. To complete two calling paths to (ra,n) requires m+n-i-j+1 time units. Theorem 3_.^9: There is a calling scheme which completes 1 fault-tolerant broad- cast from any vertex (i,j) of R with 5lj max{m+n-i-J+l, n+i-j+1, m+j-i+3, i+j+3}. This is known to be minimum time for m>2i and n>2j+2. Proof: Consider the grid R . The four subgrids R J Jf R. ,,,, R .,, ., and m,n B i,j» i,n-j+l' m-i+l,j' R -1+1 -i+i determined by the message originator (i,j) will be referred to as the upper left, upper right, lower left, and lower right subgrids respec- tively. The corner broadcast schemes used on each subgrid will assume an Page 23 orientation in which row i of the grid becomes row 1 of the subgrid and column j of the grid becomes column 1 of the subgrid. A calling scheme can be con- structed as follows: 1. Apply the 1 fault- tolerant corner broadcast scheme for R ... ... to the lower right subgrid. The call made m-i+l,n-j+l e & between (m-i+1,1) and (m-i+1,2) at time m-i+5 is delayed by one time unit. 2. Apply the 1 fault- tolerant corner broadcast scheme for R ... to the lower left subgrid. All calls except those m-i+l,j between column 1 vertices are delayed by three time units. (The times of the column 1 calls are determined by the lower right subgrid.) The second call between (m-i+1,1) and (m- i+1,2) is deleted. 3. Apply the 1 fault- tolerant corner broadcast scheme for R. _j.i to the upper right subgrid. All calls except those in row 1 are delayed by one time unit. (The times of the row 1 calls are determined by the lower right subgrid.) The second call between (i,l) and (i,2) is delayed by one addi- tional time unit. All calls which occur between row 1 and row 2 after time n-j+2 are deleted. 4. Apply the 1 fault-tolerant corner broadcast scheme for R. . to the upper left subgrid. All calls except those i » J between column 1 vertices and those between row 1 vertices are delayed by four time units. (The times of the row 1 and column 1 calls are determined by the lower left and upper right subgrids respectively.) All calls which occur between row 1 and row 2 after time j+5 or between column 1 and Page 24 column 2 after time i+5 are deleted. The scheme is illustrated by the following example. The grid is R and the message originator is (5,5). The schemes for each subgrid before delays are: 9 (2 II 12 ? /0 , n 6 . 1 lo 2 1 IZ 10 . 3 , 1 10 , '3 , e r " V l0 , H (3 JO >Z 1 5- , 10 . 12. , !.? , 13 6> 10 > *J " X , 8 . f /» // 6 9 «• II lo . ? * , 1 , '° - '1 5 6 7 1 ,^ 1 /0 i II , V 5 6 ? g 6 ■ f i . 1 J 10 3 * £• 6 ? «>-*- z 1 3 .. * ,,* , 4 a ' . *■ 3 . 4 5" y 3 5 f 4 B 5- 7" , 1 8 IO 3 It 6 ? , * , 7 Y 5" 6 9 7 , 1 to 5- 4 1 8 , 8 ID II 6 1 to it ID i lo > The time required by the above scheme is max{ t(m-i+l,n-j+l) , t(i,n- j+l)+l, t(m-i+l, j)+3, t(i,j)+4} - max{m+n-i- j+1, n+i-j+1, m+j-i+3, i+j+3} which is m+n-i-j+1 when m>2i and n>2j+2. Schemes for other interior vertices may be constructed using a similar technique. Page 26 4 Fault-tolerant Broadcasting In Wraparound Grids . Let s(m,n) [i(m,n)] be the minimum time required for 1 fault-tolerant broadcasting from a vertex of S [I ]. m,n m,n Theorem 4.1 : s(m,n) ,i(m,n) > [m/2] + [n/2] +1 if m or n is even; m,n>2 [m/2] + [n/2] if m and n are odd; m,n>3 Proof: To complete a fault-tolerant broadcast in S or in I requires r m,n m,n n [m/2 J + [n/2 J time units if m or n is even and both are greater than 2 and [m/2 J + [n/2 J + 1 time units if both m and n are odd and greater than 3 [Far- ley & Hedetniemi, 78]. Clearly any scheme which completes 2 calling paths to each vertex of either grid must take at least one time unit more than a fault-tolerant scheme. Theorem j4.2_: There is a calling scheme which completes 1 fault-tolerant broad- cast from any vertex of S or I with m,n>4 in [m/2] + [n/2l + 1 time m,n m,n ' L J L J units. This is known to be minimum time if m or n is even. Proof : Consider S . Without loss of generality m>n and the message origina- tor Is ((m/2), (n/2) ). Let p=(m/2) and q=(n/2). The following scheme com- pletes 1 fault-tolerant broadcast in [m/2j + [n/2] + 1 time units: vertex time talks to vertex time talks to (1,1) p+q p+q+1 (2,1) (l,n) (1,2) p+q p+q+1 (2,2) (1.3) Cl.J) 2n. This scheme may be applied directly to I if n>m. That is, after reorienting I the 3 KK J m,n ' 6 m,n scheme may be used since it ignores the edges between rows 1 and n. If n4 or on S. or I. with n>4 in minimum time. m,4 m,4 4,n 4,n Proof ; Consider S ,. Without loss of generality the message originator is 1 III y H (fm/2),2). The following scheme informs all vertices of S . with m>4 by time m, 4 [m/2J + 3 which is minimum by Theorem 4.1: a fl+Z a»3 4tl At3 H*l c a . ai-3 '. 3 3 Q- \p+-l a V>H to+l k>+Z 7)a+3 D^ti This scheme may also be used on S, or on I, by reorienting them as above. In S, ,, six time units are required to complete 1 fault-tolerant broad- cast. This is easily seen since all informed vertices must be involved in calls at times 1 through 4 in order to complete a single path to every vertex by time 4. In particular the message originator must call each of its neigh- bors. Thus at time 5 no second paths can be completed by a call from the mes- sage originator. Only an even number of second paths can be completed at time 5 since each call completes two of them. Since 15 points require second paths Page 31 an additional call must be made at time 6 to complete the scheme. The follow- ing scheme informs all vertices of S, , by t»6: CH c; c; .^ZD Z)5 5 In order to use these schemes for I , the following changes must be made: 1. Edges between (1+1, 1 ) and (i,n) with Ki3 or S or I- with m>3 in m,J m,J J,m J,m time [m/2j + 3 if m is odd or [m/2] + 4 if m is even. Proof : Consider S .. Without loss of generality the message originator is Page 32 (fm/2),2). The following schemes complete the proof for S .: m, J CZK^ p* .+■>. l+l i+T- a-H c i *** i J_J IO.+ 2. (■.«*» l«.+ i c: a-»-3 «.rX ( J* U af2 , ar3 , r A** r , at3 J a 1 *±2-h 4+3 , a+4 a+L <*-2 *H . a+1 ,, d-rt , a+3 a+4 '. 4 H 3 c . z r 1 | * 3 < 3 2 . ao . *H a+3 a+H at-Z +3 a = /n oM <*+3 at-f m even As above these schemes may be applied directly to S and I- . For 1 the J,m J,m m, j following changes must be made: 1. If m is odd label the edges between (i+1,1) and (i,n) for Ki<[m/2j+2. 2. If m is even label the edges ([m/2j,l), ([m/2J-l,n) and ([m/2),n), ([m/2]+l,l) with [m/2]+4. Corollary f*_»b} The 1 fault-tolerant scheme above requires mn+2m total calls as compared to mn-1 for the fault-tolerant scheme. It should be noted that the wraparound grids are A edge-connected and will admit 2 fault-tolerant and 3 fault-tolerant broadcast schemes [Liestman, 80]. Such schemes would include three and four calling paths to each vertex. Page 33 5 Summary * The process of fault-tolerant broadcasting has been defined and investi- gated in grid and grid-like graphs. The results show that the addition of 1 fault-tolerance to the broadcasting process requires a small increase in the time required for broadcasting and a substantial increase in system utiliza- tion. In particular, only a single time unit beyond the fault-tolerant broadcast time is required to perform a 1 fault-tolerant broadcast on most grids. The total number of calls required is roughly doubled when fault- tolerance is added. Page 34 References. [Cockayne & Hedetniemi, 78] Cockayne, E. J. and S. T. Hedetniemi, "A Conjec- ture Concerning Broadcasting in M-dimensional Grid Graphs", Univ. of Ore- gon Tech. Rept. No. CS-TR-78-14, 1978. [Farley, 79] Farley, A. M. , "Minimal Broadcast Networks", Networks, Vol. 9, No. 4, pp. 313-332, 1979. [Farley & Hedetniemi, 78] Farley, A. M. and S. T. Hedetniemi, "Broadcasting in Grid Graphs", Univ. of Oregon Tech. Rept. No. CS-TR-78-6, 1978. [Farley, et al. , 79] Farley, A. M. , S. T. Hedetniemi, S. Mitchell and A. Proskurowski , "Minimum Broadcast Graphs", Discrete Mathematics, Vol. 25, pp. 189-193, 1979. [Harary, 69] Harary, F. , Graph Theory , Addison-Wesley, Reading, Mass., 1969. [Liestman, 80] Liestman, A., "Fault-tolerant Broadcast Graphs", Univ. of Illi- nois Tech. Rept. No. UIUCDCS-R-80-1029, 1980. [Mitchell & Hedetniemi, 78] Mitchell, S. L. and S. T. Hedetniemi, "A Census of Minimum Broadcast Graphs", Univ. of Oregon Tech. Rept. No. CS-TR-78-7, 1978. [Peck, 80] Peck, G. W. , "Optimal Spreading in an n Dimensional Rectilinear Grid", Studies in Applied Mathematics, Vol. 62, No. 1, pp. 69-74, 1980. [Van Scoy, 79] Van Scoy, F. L. , "Broadcasting a Small Number of Messages in a Square Grid Graph", Proceedings of the Seventeenth Annual Allerton Conference on Communication, Control and Computing, 1979. BIBLIOGRAPHIC DATA SHEET 1. Report No. UIUCDCS-R-80-1030 3. Recipient's Accession No. 5. Report Date 4. Title and Subtitle FAULT-TOLERANT GRID BROADCASTING 6. 7. Author(s) Arthur L. Liestman 8. Performing Organization Rept. No, R-80-1030 9. Performing Organization Name and Address Department of Computer Science University of Illinois U-C Urbana, IL 61801 10. Project/Task/Work Unit No. 11. Contract /Grant No. NSG 1471 12. Sponsoring Organization Name and Address National Aeronautics and Space Administration Hampton, Virginia 13. Type of Report & Period Covered Technical 14. 15. Supplementary Notes 16. Abstracts Fault-tolerant broadcasting is simply broadcasting with redundancy in order that certain faults, i.e. connections between nodes ceasing to function, do not affect the completion of the broadcast. This paper investigates the problem of fault-tolerant broadcasting in grid or grid-like graphs. 17. Key Words and Document Analysis. 17a. Descriptors fault- tolerant broadcasting grids 17b. Identifiers /Open-Ended Terms 17c. COSATI Field/Group 18. Availability Statement unlimited 19. Security Class (This Report) UNCLASSIFIED 20. Security Class (This Page UNCLASSIFIED 21. No. of Pages 37 22. Price FORM NTIS-35 (10-70) USCOMM-DC 40329-P71