add option to print latency distribution

src/stats.c

@@ -23,27 +23,21 @@ void stats_record(stats *stats, uint64_t x) {
     if (stats->index == stats->samples) stats->index = 0;
 }
 
-uint64_t stats_min(stats *stats) {
-    uint64_t min = 0;
-    for (uint64_t i = 0; i < stats->limit; i++) {
-        uint64_t x = stats->data[i];
-        if (x < min || min == 0) min = x;
-    }
-    return min;
+static int stats_compare(const void *a, const void *b) {
+    uint64_t *x = (uint64_t *) a;
+    uint64_t *y = (uint64_t *) b;
+    return *x - *y;
 }
 
-uint64_t stats_max(stats *stats) {
-    uint64_t max = 0;
-    for (uint64_t i = 0; i < stats->limit; i++) {
-        uint64_t x = stats->data[i];
-        if (x > max || max == 0) max = x;
-    }
-    return max;
-}
+long double stats_summarize(stats *stats, int64_t *min, uint64_t *max) {
+    qsort(stats->data, stats->limit, sizeof(uint64_t), &stats_compare);
+
+    if (min) *min = stats->data[0];
+    if (max) *max = stats->data[stats->limit - 1];
 
-long double stats_mean(stats *stats) {
-    uint64_t sum = 0;
     if (stats->limit == 0) return 0.0;
+
+    uint64_t sum = 0;
     for (uint64_t i = 0; i < stats->limit; i++) {
         sum += stats->data[i];
     }
@@ -71,3 +65,8 @@ long double stats_within_stdev(stats *stats, long double mean, long double stdev
 
     return (sum / (long double) stats->limit) * 100;
 }
+
+uint64_t stats_percentile(stats *stats, long double p) {
+    uint64_t rank = round((p / 100.0) * stats->limit + 0.5);
+    return stats->data[rank - 1];
+}

src/stats.h

@@ -11,11 +11,10 @@ typedef struct {
 stats *stats_alloc(uint64_t);
 void stats_free(stats *);
 void stats_record(stats *, uint64_t);
-uint64_t stats_min(stats *);
-uint64_t stats_max(stats *);
-long double stats_mean(stats *);
+long double stats_summarize(stats *, int64_t *, uint64_t *);
 long double stats_stdev(stats *stats, long double);
 long double stats_within_stdev(stats *, long double, long double, uint64_t);
+uint64_t stats_percentile(stats *, long double);
 
 #endif /* STATS_H */
 

src/wrk.c

@@ -11,6 +11,7 @@
 #include <netinet/in.h>
 #include <netinet/tcp.h>
 #include <stdarg.h>
+#include <stdbool.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -33,6 +34,7 @@ static struct config {
     uint64_t connections;
     uint64_t duration;
     uint64_t timeout;
+    bool     latency;
 } cfg;
 
 static struct {
@@ -64,6 +66,7 @@ static void usage() {
            "    -t, --threads     <N>  Number of threads to use   \n"
            "                                                      \n"
            "    -H, --header      <H>  Add header to request      \n"
+           "        --latency          Print latency statistics   \n"
            "        --timeout     <T>  Socket/request timeout     \n"
            "    -v, --version          Print version details      \n"
            "                                                      \n"
@@ -187,6 +190,7 @@ int main(int argc, char **argv) {
     print_stats_header();
     print_stats("Latency", statistics.latency, format_time_us);
     print_stats("Req/Sec", statistics.requests, format_metric);
+    if (cfg.latency) print_stats_latency(statistics.latency);
 
     char *runtime_msg = format_time_us(runtime_us);
 
@@ -428,6 +432,7 @@ static struct option longopts[] = {
     { "duration",    required_argument, NULL, 'd' },
     { "threads",     required_argument, NULL, 't' },
     { "header",      required_argument, NULL, 'H' },
+    { "latency",     no_argument,       NULL, 'L' },
     { "timeout",     required_argument, NULL, 'T' },
     { "help",        no_argument,       NULL, 'h' },
     { "version",     no_argument,       NULL, 'v' },
@@ -443,7 +448,7 @@ static int parse_args(struct config *cfg, char **url, char **headers, int argc,
     cfg->duration    = 10;
     cfg->timeout     = SOCKET_TIMEOUT_MS;
 
-    while ((c = getopt_long(argc, argv, "t:c:d:H:T:rv?", longopts, NULL)) != -1) {
+    while ((c = getopt_long(argc, argv, "t:c:d:H:T:Lrv?", longopts, NULL)) != -1) {
         switch (c) {
             case 't':
                 if (scan_metric(optarg, &cfg->threads)) return -1;
@@ -457,6 +462,9 @@ static int parse_args(struct config *cfg, char **url, char **headers, int argc,
             case 'H':
                 *header++ = optarg;
                 break;
+            case 'L':
+                cfg->latency = true;
+                break;
             case 'T':
                 if (scan_time(optarg, &cfg->timeout)) return -1;
                 cfg->timeout *= 1000;
@@ -506,8 +514,8 @@ static void print_units(long double n, char *(*fmt)(long double), int width) {
 }
 
 static void print_stats(char *name, stats *stats, char *(*fmt)(long double)) {
-    long double mean  = stats_mean(stats);
-    long double max   = stats_max(stats);
+    uint64_t max;
+    long double mean  = stats_summarize(stats, NULL, &max);
     long double stdev = stats_stdev(stats, mean);
 
     printf("    %-10s", name);
@@ -516,3 +524,15 @@ static void print_stats(char *name, stats *stats, char *(*fmt)(long double)) {
     print_units(max,   fmt, 9);
     printf("%8.2Lf%%\n", stats_within_stdev(stats, mean, stdev, 1));
 }
+
+static void print_stats_latency(stats *stats) {
+    long double percentiles[] = { 50.0, 75.0, 90.0, 99.0 };
+    printf("  Latency Distribution\n");
+    for (size_t i = 0; i < sizeof(percentiles) / sizeof(long double); i++) {
+        long double p = percentiles[i];
+        uint64_t n = stats_percentile(stats, p);
+        printf("%7.0Lf%%", p);
+        print_units(n, format_time_us, 10);
+        printf("\n");
+    }
+}

src/wrk.h

@@ -71,5 +71,6 @@ static char *format_request(char *, char *, char *, char **);
 static int parse_args(struct config *, char **, char **, int, char **);
 static void print_stats_header();
 static void print_stats(char *, stats *, char *(*)(long double));
+static void print_stats_latency(stats *);
 
 #endif /* WRK_H */