1 | // Frequent (>1000) failure cause for different categories of events grouped by duration (short, medium, long) |
---|
2 | |
---|
3 | #include <stdio.h> |
---|
4 | #include <stdlib.h> |
---|
5 | #include <string.h> |
---|
6 | #include "mpi.h" |
---|
7 | |
---|
8 | #define CHUNK 20480 |
---|
9 | #define ENCODED_NO 33 |
---|
10 | |
---|
11 | |
---|
12 | unsigned long* poz; |
---|
13 | unsigned long recv[2]; |
---|
14 | unsigned long poz_size = 0; |
---|
15 | unsigned long recv_m[ENCODED_NO]; |
---|
16 | unsigned long send_m[ENCODED_NO]; |
---|
17 | |
---|
18 | int failure_key[] = {-1, 0, 999, 1999, 2999, 3999, 4999, 5999, 6999, 7000, 7001}; |
---|
19 | char* failure_value[] = {"not reported", "reported as undetermined", "infrastructure", "hardware", "IO", "network", "software", "human error", |
---|
20 | "user", "end of measurement", "TYPING"}; |
---|
21 | long duration_key[] = {1000, 100000, 2147483647}; |
---|
22 | char* duration_value[] = {"short", "medium", "long"}; |
---|
23 | char* encoded_values[] = {"short-not reported", "medium-not reported", "long-not reported", "short-reported as undetermined", |
---|
24 | "medium-reported as undetermined", "long-reported as undetermined", "short-infrastructure", "medium-infrastructure", |
---|
25 | "long-infrastructure", "short-hardware", "medium-hardware", "long-hardware", "short-IO", "medium-IO", "long-IO", "short-network", |
---|
26 | "medium-network", "long-network", "short-software", "medium-software", "long-software", "short-human error", |
---|
27 | "medium-human error", "long-human error", "short-user", "medium-user", "long-user", "short-end of measurement", |
---|
28 | "medium-end of measurement", "long-end of measurement", "short-TYPING", "medium-TYPING", "long-TYPING"}; |
---|
29 | |
---|
30 | unsigned long file_length(FILE *f) { |
---|
31 | long pos; |
---|
32 | long end; |
---|
33 | |
---|
34 | pos = ftell (f); |
---|
35 | fseek (f, 0, SEEK_END); |
---|
36 | end = ftell (f); |
---|
37 | fseek (f, pos, SEEK_SET); |
---|
38 | |
---|
39 | return end; |
---|
40 | } |
---|
41 | |
---|
42 | |
---|
43 | void write_result (char *filename) { |
---|
44 | FILE *f; |
---|
45 | int i; |
---|
46 | |
---|
47 | f = fopen(filename, "wt"); |
---|
48 | if (f == NULL) { |
---|
49 | fprintf(stderr, "Could not open file write.\n"); |
---|
50 | MPI_Abort(MPI_COMM_WORLD, 7777); |
---|
51 | } |
---|
52 | |
---|
53 | for(i = 0 ; i < ENCODED_NO ; i++) { |
---|
54 | fprintf(f, "%-40s", encoded_values[i]); |
---|
55 | fprintf(f, "%i\n", recv_m[i]); |
---|
56 | } |
---|
57 | } |
---|
58 | |
---|
59 | int init(char *filename) { |
---|
60 | unsigned long file_len = 0; |
---|
61 | unsigned long i, j; |
---|
62 | FILE *f; |
---|
63 | size_t read = 0; |
---|
64 | |
---|
65 | f = fopen(filename, "rt"); |
---|
66 | if (f == NULL) { |
---|
67 | fprintf(stderr, "Could not open file init.\n"); |
---|
68 | MPI_Abort(MPI_COMM_WORLD, 7777); |
---|
69 | } |
---|
70 | file_len = file_length(f); |
---|
71 | poz_size = file_len / CHUNK + 1; |
---|
72 | |
---|
73 | poz = (unsigned long*) malloc(poz_size * sizeof(unsigned long)); |
---|
74 | poz[0] = 0; |
---|
75 | for (i = CHUNK, j = 1; j < poz_size - 1; i += CHUNK, j++) { |
---|
76 | poz[j] = i; |
---|
77 | } |
---|
78 | poz[j] = file_len; |
---|
79 | |
---|
80 | return 1; |
---|
81 | } |
---|
82 | |
---|
83 | int main(int argc, char** argv) { |
---|
84 | |
---|
85 | int rank, size; |
---|
86 | MPI_Status status; |
---|
87 | MPI_Request request; |
---|
88 | int i, flag, j; |
---|
89 | |
---|
90 | MPI_Init(&argc, &argv); |
---|
91 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); |
---|
92 | MPI_Comm_size(MPI_COMM_WORLD, &size); |
---|
93 | |
---|
94 | if (rank == 0) { |
---|
95 | // master inits |
---|
96 | init(argv[1]); |
---|
97 | } |
---|
98 | MPI_Barrier(MPI_COMM_WORLD); |
---|
99 | |
---|
100 | // master sends indexes, receives results |
---|
101 | if(rank == 0) { |
---|
102 | |
---|
103 | for(i = 0 ; i < ENCODED_NO ; i++) { |
---|
104 | recv_m[i] = 0; |
---|
105 | send_m[i] = 0; |
---|
106 | } |
---|
107 | |
---|
108 | // sends indexes |
---|
109 | for (i = 0; i < poz_size - 1; i++) { |
---|
110 | MPI_Isend(poz+i, 2, MPI_UNSIGNED_LONG, i%(size-1)+1, 0, MPI_COMM_WORLD, &request); |
---|
111 | } |
---|
112 | |
---|
113 | // receive results |
---|
114 | unsigned long recv_temp[ENCODED_NO]; |
---|
115 | while(1) { |
---|
116 | |
---|
117 | for(i = 0 ; i < ENCODED_NO ; i++) |
---|
118 | recv_temp[i] = 0; |
---|
119 | MPI_Irecv(recv_temp, ENCODED_NO, MPI_UNSIGNED_LONG, MPI_ANY_SOURCE, 0, MPI_COMM_WORLD, &request); |
---|
120 | |
---|
121 | MPI_Test(&request, &flag, &status); |
---|
122 | int times = 0; |
---|
123 | while(!flag && times < 10) { |
---|
124 | times++; |
---|
125 | sleep(1); |
---|
126 | MPI_Test(&request, &flag, &status); |
---|
127 | } |
---|
128 | if(!flag) |
---|
129 | break; |
---|
130 | |
---|
131 | for(i = 0 ; i < ENCODED_NO ; i++) { |
---|
132 | recv_m[i] += recv_temp[i]; |
---|
133 | } |
---|
134 | } |
---|
135 | |
---|
136 | // write results |
---|
137 | write_result(argv[2]); |
---|
138 | free(poz); |
---|
139 | } |
---|
140 | |
---|
141 | // slaves receive indexes, do computation |
---|
142 | else { |
---|
143 | while(1) { |
---|
144 | |
---|
145 | // receive until there's nothing left |
---|
146 | MPI_Irecv(recv, 2, MPI_UNSIGNED_LONG, 0, 0, MPI_COMM_WORLD, &request); |
---|
147 | |
---|
148 | MPI_Test(&request, &flag, &status); |
---|
149 | int times = 0; |
---|
150 | while(!flag && times < 10) { |
---|
151 | times++; |
---|
152 | sleep(1); |
---|
153 | MPI_Test(&request, &flag, &status); |
---|
154 | } |
---|
155 | if(!flag) |
---|
156 | break; |
---|
157 | |
---|
158 | // read file and do computation |
---|
159 | FILE *f; |
---|
160 | |
---|
161 | f = fopen(argv[1], "rt"); |
---|
162 | if (f == NULL) { |
---|
163 | fprintf(stderr, "Could not open file.\n"); |
---|
164 | MPI_Abort(MPI_COMM_WORLD, 7777); |
---|
165 | } |
---|
166 | |
---|
167 | fseek (f, recv[0], SEEK_SET); |
---|
168 | |
---|
169 | // go to the end of the line |
---|
170 | char * line = NULL; |
---|
171 | size_t len = 0; |
---|
172 | ssize_t read; |
---|
173 | unsigned long count = 0; |
---|
174 | |
---|
175 | for(i = 0 ; i < ENCODED_NO ; i++) { |
---|
176 | send_m[i] = 0; |
---|
177 | recv_m[i] = 0; |
---|
178 | } |
---|
179 | |
---|
180 | while ((read = getline(&line, &len, f)) != -1) { |
---|
181 | |
---|
182 | // skip the comment line and the half lines |
---|
183 | count++; |
---|
184 | if(count == 1) |
---|
185 | continue; |
---|
186 | |
---|
187 | // compute |
---|
188 | char response; |
---|
189 | |
---|
190 | double start, end, duration; |
---|
191 | int fault = -2; |
---|
192 | char delims[] = " "; |
---|
193 | char *result = NULL; |
---|
194 | result = strtok( line, delims ); |
---|
195 | int token_no = 0; |
---|
196 | while( result != NULL ) { |
---|
197 | token_no++; |
---|
198 | if(token_no == 7) { |
---|
199 | start = atof(result); |
---|
200 | } |
---|
201 | if(token_no == 8) { |
---|
202 | end = atof(result); |
---|
203 | } |
---|
204 | if(token_no == 9) { |
---|
205 | if(strncmp(result, "NULL", 4) == 0) |
---|
206 | fault = -1; |
---|
207 | else |
---|
208 | fault = atoi(result); |
---|
209 | } |
---|
210 | result = strtok( NULL, delims ); |
---|
211 | } |
---|
212 | |
---|
213 | int tag = 0; |
---|
214 | |
---|
215 | if(fault > failure_key[10]) { |
---|
216 | // detect typing fault |
---|
217 | fault = 7001; |
---|
218 | } |
---|
219 | for(i = 0 ; i < 11 ; i++) { |
---|
220 | if(fault <= failure_key[i]) { |
---|
221 | tag = i * 3; |
---|
222 | break; |
---|
223 | } |
---|
224 | } |
---|
225 | |
---|
226 | duration = end - start; |
---|
227 | for(i = 0 ; i < 3 ; i++) { |
---|
228 | if(duration <= duration_key[i]) { |
---|
229 | tag += i; |
---|
230 | break; |
---|
231 | } |
---|
232 | } |
---|
233 | |
---|
234 | send_m[tag]++; |
---|
235 | |
---|
236 | // finished reading chunk? |
---|
237 | if(ftell(f) > recv[1]) |
---|
238 | break; |
---|
239 | |
---|
240 | } |
---|
241 | |
---|
242 | // send result to master |
---|
243 | MPI_Isend(send_m, ENCODED_NO, MPI_UNSIGNED_LONG, 0, 0, MPI_COMM_WORLD, &request); |
---|
244 | } |
---|
245 | } |
---|
246 | |
---|
247 | // block and finish |
---|
248 | MPI_Barrier(MPI_COMM_WORLD); |
---|
249 | MPI_Finalize(); |
---|
250 | } |
---|
251 | |
---|