Confirm that the rrd file is writable before starting

Check time period vs send interval and loss interval to ensure there is always one resolved slot

dpinger.c

@@ -74,16 +74,17 @@ static const char *             pidfile_name = NULL;
 // Flags
 static unsigned int             flag_rewind = 0;
 static unsigned int             flag_syslog = 0;
+static unsigned int             flag_priority = 0;
 
 // String representation of target
 #define ADDR_STR_MAX            (INET6_ADDRSTRLEN + IF_NAMESIZE + 1)
 static char                     dest_str[ADDR_STR_MAX];
 
 // Time period over which we are averaging results in ms
-static unsigned long            time_period_msec = 30000;
+static unsigned long            time_period_msec = 60000;
 
 // Interval between sends in ms
-static unsigned long            send_interval_msec = 250;
+static unsigned long            send_interval_msec = 500;
 
 // Interval before a sequence is initially treated as lost
 // Input from command line in ms and used in us
@@ -286,7 +287,7 @@ llsqrt(
         }
     }
 
-    return s;
+    return (unsigned long) s;
 }
 
 
@@ -352,14 +353,14 @@ send_thread(
 
         array[next_slot].status = PACKET_STATUS_EMPTY;
         sched_yield();
-        clock_gettime(CLOCK_MONOTONIC, &array[next_slot].time_sent);
 
+        clock_gettime(CLOCK_MONOTONIC, &array[next_slot].time_sent);
+        array[next_slot].status = PACKET_STATUS_SENT;
         len = sendto(send_sock, echo_request, echo_request_len, 0, (struct sockaddr *) &dest_addr, dest_addr_len);
         if (len == -1)
         {
             logger("%s%s: sendto error: %d\n", identifier, dest_str, errno);
         }
-        array[next_slot].status = PACKET_STATUS_SENT;
 
         next_slot = (next_slot + 1) % array_size;
         next_sequence = (next_sequence + 1) % sequence_limit;
@@ -471,7 +472,7 @@ report(
             packets_received++;
             latency_usec = array[slot].latency_usec;
             total_latency_usec += latency_usec;
-            total_latency_usec2 += latency_usec * latency_usec;
+            total_latency_usec2 += (unsigned long long) latency_usec * latency_usec;
         }
         else if (array[slot].status == PACKET_STATUS_SENT &&
                  ts_elapsed_usec(&array[slot].time_sent, &now) > loss_interval_usec)
@@ -489,7 +490,7 @@ report(
 
         // stddev = sqrt((sum(rtt^2) / packets) - (sum(rtt) / packets)^2)
         *average_latency_usec = avg;
-        *latency_deviation = llsqrt(avg2 - (avg * avg));
+        *latency_deviation = llsqrt(avg2 - ((unsigned long long) avg * avg));
     }
     else
     {
@@ -824,15 +825,16 @@ static void
 usage(void)
 {
     fprintf(stderr, "Usage:\n");
-    fprintf(stderr, "  %s [-f] [-R] [-S] [-B bind_addr] [-s send_interval] [-l loss_interval] [-t time_period] [-r report_interval] [-d data_length] [-o output_file] [-A alert_interval] [-D latency_alarm] [-L loss_alarm] [-C alert_cmd] [-i identifier] [-u usocket] [-p pidfile] dest_addr\n\n", progname);
+    fprintf(stderr, "  %s [-f] [-R] [-S] [-P] [-B bind_addr] [-s send_interval] [-l loss_interval] [-t time_period] [-r report_interval] [-d data_length] [-o output_file] [-A alert_interval] [-D latency_alarm] [-L loss_alarm] [-C alert_cmd] [-i identifier] [-u usocket] [-p pidfile] dest_addr\n\n", progname);
     fprintf(stderr, "  options:\n");
     fprintf(stderr, "    -f run in foreground\n");
     fprintf(stderr, "    -R rewind output file between reports\n");
     fprintf(stderr, "    -S log warnings via syslog\n");
+    fprintf(stderr, "    -P priority scheduling for receive thread (requires root)\n");
     fprintf(stderr, "    -B bind (source) address\n");
-    fprintf(stderr, "    -s time interval between echo requests (default 250ms)\n");
-    fprintf(stderr, "    -l time interval before packets are treated as lost (default 5x send interval)\n");
-    fprintf(stderr, "    -t time period over which results are averaged (default 30s)\n");
+    fprintf(stderr, "    -s time interval between echo requests (default 500ms)\n");
+    fprintf(stderr, "    -l time interval before packets are treated as lost (default 4x send interval)\n");
+    fprintf(stderr, "    -t time period over which results are averaged (default 60s)\n");
     fprintf(stderr, "    -r time interval between reports (default 1s)\n");
     fprintf(stderr, "    -d data length (default 0)\n");
     fprintf(stderr, "    -o output file for reports (default stdout)\n");
@@ -866,14 +868,11 @@ fatal(
     const char *                format,
     ...)
 {
-    if (format)
-    {
-        va_list                 args;
+    va_list                 args;
 
-        va_start(args, format);
-        vfprintf(stderr, format, args);
-        va_end(args);
-    }
+    va_start(args, format);
+    vfprintf(stderr, format, args);
+    va_end(args);
 
     exit(EXIT_FAILURE);
 }
@@ -897,7 +896,7 @@ parse_args(
 
     progname = argv[0];
 
-    while((opt = getopt(argc, argv, "fRSB:s:l:t:r:d:o:A:D:L:C:i:u:p:")) != -1)
+    while((opt = getopt(argc, argv, "fRSPB:s:l:t:r:d:o:A:D:L:C:i:u:p:")) != -1)
     {
         switch (opt)
         {
@@ -913,6 +912,10 @@ parse_args(
             flag_syslog = 1;
             break;
 
+        case 'P':
+            flag_priority = 1;
+            break;
+
         case 'B':
             bind_arg = optarg;
             break;
@@ -1018,7 +1021,7 @@ parse_args(
 
         default:
             usage();
-            fatal(NULL);
+            exit(EXIT_FAILURE);
         }
     }
 
@@ -1026,7 +1029,7 @@ parse_args(
     if (argc != optind + 1)
     {
         usage();
-        fatal(NULL);
+        exit(EXIT_FAILURE);
     }
     dest_arg = argv[optind];
 
@@ -1036,17 +1039,17 @@ parse_args(
         fatal("no activity enabled\n");
     }
 
-    // Ensure we have something to average over
-    if (time_period_msec < send_interval_msec)
+    // Ensure there is a minimum of one resolved slot at all times
+    if (time_period_msec <= send_interval_msec * 2 + loss_interval_msec)
     {
-        fatal("time period cannot be less than send interval\n");
+        fatal("the time period must be greater than twice the send interval plus the loss interval\n");
     }
 
     // Ensure we don't have sequence space issues. This really should only be hit by
     // complete accident. Even a ratio of 16384:1 would be excessive.
     if (time_period_msec / send_interval_msec > 65536)
     {
-        fatal("ratio of time period to send interval cannot exceed 65536:1\n");
+        fatal("the ratio of time period to send interval cannot exceed 65536:1\n");
     }
 
     // Check destination address
@@ -1320,7 +1323,7 @@ main(
     // Set the default loss interval
     if (loss_interval_msec == 0)
     {
-        loss_interval_msec = send_interval_msec * 5;
+        loss_interval_msec = send_interval_msec * 4;
     }
     loss_interval_usec = loss_interval_msec * 1000;
 
@@ -1363,6 +1366,27 @@ main(
         fatal("cannot create recv thread\n");
     }
 
+    // Set priority on recv thread if requested
+    if (flag_priority)
+    {
+        struct sched_param          thread_sched_param;
+
+        r = sched_get_priority_min(SCHED_RR);
+        if (r == -1)
+        {
+            perror("sched_get_priority_min");
+            fatal("cannot determin minimum shceduling priority for SCHED_RR\n");
+        }
+        thread_sched_param.sched_priority = r;
+
+        r = pthread_setschedparam(thread, SCHED_RR, &thread_sched_param);
+        if (r != 0)
+        {
+            perror("pthread_setschedparam");
+            fatal("cannot set receive thread priority\n");
+        }
+    }
+
     // Create send thread
     r = pthread_create(&thread, NULL, &send_thread, NULL);
     if (r != 0)

rrd/dpinger_rrd_create

@@ -10,10 +10,21 @@ name="$1"
 rrdfile="${name}.rrd"
 echo "Creating rrd file ${rrdfile}"
 
-rrdtool create "${rrdfile}" --step 1m \
-    DS:latency:GAUGE:5m:0:U \
-    DS:stddev:GAUGE:5m:0:U \
-    DS:loss:GAUGE:5m:0:100 \
-    RRA:AVERAGE:0.5:1m:15d \
-    RRA:AVERAGE:0.5:5m:90d \
-    RRA:AVERAGE:0.5:1h:3y
+
+# Time duration method doesn't work in all versions of rrdtool
+#rrdtool create "${rrdfile}" --step 1m \
+#    DS:latency:GAUGE:5m:0:U \
+#    DS:stddev:GAUGE:5m:0:U \
+#    DS:loss:GAUGE:5m:0:100 \
+#    RRA:AVERAGE:0.5:1m:15d \
+#    RRA:AVERAGE:0.5:5m:90d \
+#    RRA:AVERAGE:0.5:1h:3y
+
+# This method works in all versions
+rrdtool create "${rrdfile}" --step 60 \
+    DS:latency:GAUGE:300:0:U \
+    DS:stddev:GAUGE:300:0:U \
+    DS:loss:GAUGE:300:0:100 \
+    RRA:AVERAGE:0.5:1:21600 \
+    RRA:AVERAGE:0.5:5:25920 \
+    RRA:AVERAGE:0.5:60:26352

rrd/dpinger_rrd_update

@@ -15,6 +15,11 @@ dpinger=/usr/local/bin/dpinger
 
 
 rrdfile="${name}.rrd"
+if [ \! -w ${rrdfile} ]
+then
+    echo "$0: file \"${rrdfile}\" does not exist or is not writable"
+    exit 1
+fi
 
 ${dpinger} -f ${options} -s 500m -t 60s -r 60s ${targetip} |
 while read -r latency stddev loss; do

rrd/sample.html

@@ -1,5 +1,5 @@
 <html>
-    <head><title>WAN Statistics</title></head>
+    <head><title>>Latency Statistics for WAN</title></head>
     <body>
         <img src="/tmp/wan-1.png" alt="wan-1">
         <p>