[TESTS] add new methods in ip-hash test file

added methods to provide a better hash with small input sets

diff --git a/tests/ip-hash.c b/tests/ip-hash.c
index 2ffd9a2cc902e303c441f70264bf2be4f8432bb2..8bb2d488fc424e418516ad495b2991c0648cc0fe 100644 (file)
--- a/tests/ip-hash.c
+++ b/tests/ip-hash.c
@@ -84,6 +84,7 @@ uint32_t hash_tw3(uint32_t a)
  * About 0x50 bytes, 6 shifts.
  */
 int counts_bj6[NSERV][NSERV];
+int counts_bj6x[NSERV][NSERV];
 uint32_t hash_bj6(uint32_t a)
 {
        a = (a+0x7ed55d16) + (a<<12);
@@ -100,6 +101,7 @@ uint32_t hash_bj6(uint32_t a)
  * About 0x40 bytes, 7 shifts.
  */
 int counts_bj7[NSERV][NSERV];
+int counts_bj7x[NSERV][NSERV];
 uint32_t hash_bj7(uint32_t a)
 {
        a -= (a<<6);
@@ -134,7 +136,8 @@ void dump_hash_results(char *name, int counts[NSERV][NSERV]) {
                for (srv = 0; srv <= nsrv; srv++) {
                        err = 100.0*(counts[nsrv][srv] - (double)counts[0][0]/(nsrv+1)) / (double)counts[0][0];
                        //printf("%6d ", counts[nsrv][srv]);
-                       printf("% 3.1f%% ", err);
+                       printf("% 3.1f%%%c ", err,
+                              counts[nsrv][srv]?' ':'*');  /* display '*' when a server is never selected */
                        err = fabs(err);
                        total_err += err;
                        if (err > max_err)
@@ -142,7 +145,7 @@ void dump_hash_results(char *name, int counts[NSERV][NSERV]) {
                }
                total_err /= (double)(nsrv+1);
                for (srv = nsrv+1; srv < NSERV; srv++)
-                       printf("      ");
+                       printf("       ");
                printf("  avg_err=%3.1f, max_err=%3.1f\n", total_err, max_err);
        }
        printf("\n");
@@ -160,10 +163,13 @@ int main() {
        memset(counts_bj6, 0, sizeof(counts_bj6));
        memset(counts_bj7, 0, sizeof(counts_bj7));
 
-       mask = 0xFFF00F00;  // user mask to apply to addresses
-       for (nr = 0; nr < 100000; nr++) {
-               address += ~nr;  // semi-random addresses.
+       address = 0x10000000;
+       mask = 0xffffff00;  // user mask to apply to addresses
+       for (nr = 0; nr < 0x10; nr++) {
+               //address += ~nr;  // semi-random addresses.
                //address += 1;
+               address += 0x00000100;
+               //address += 0x11111111;
                //address += 7;
                //address += 8;
                //address += 256;
@@ -176,6 +182,12 @@ int main() {
                count_hash_results(hash_tw3(address & mask), counts_tw3);  // 1.01s / 100M
                count_hash_results(hash_bj6(address & mask), counts_bj6);  // 1.07s / 100M
                count_hash_results(hash_bj7(address & mask), counts_bj7);  // 1.20s / 100M
+               /* adding the original address after the hash reduces the error
+                * rate in in presence of very small data sets (eg: 16 source
+                * addresses for 8 servers). In this case, bj7 is very good.
+                */
+               count_hash_results(hash_bj6(address & mask)+(address&mask), counts_bj6x); // 1.07s / 100M
+               count_hash_results(hash_bj7(address & mask)+(address&mask), counts_bj7x); // 1.20s / 100M
        }
 
        dump_hash_results("hash_id", counts_id);
@@ -183,6 +195,8 @@ int main() {
        dump_hash_results("hash_tw2", counts_tw2);
        dump_hash_results("hash_tw3", counts_tw3);
        dump_hash_results("hash_bj6", counts_bj6);
+       dump_hash_results("hash_bj6x", counts_bj6x);
        dump_hash_results("hash_bj7", counts_bj7);
+       dump_hash_results("hash_bj7x", counts_bj7x);
        return 0;
 }