diff --git a/benchmarks/benchmark.pl b/benchmarks/benchmark.pl
index 8b135b3..62580d0 100755
--- a/benchmarks/benchmark.pl
+++ b/benchmarks/benchmark.pl
@@ -46,7 +46,7 @@ sub mk_rsock {
 }
 
 sub send_benchmark {
-    my ($lib, $addr, $port, $iter, $threads, $gen_recv, @fps_vals) = @_;
+    my ($lib, $addr, $port, $iter, $threads, $gen_recv, $mode, @fps_vals) = @_;
     my ($socket, $remote, $data);
 
     $socket = mk_ssock($addr, $port);
@@ -59,7 +59,7 @@ sub send_benchmark {
 
             for ((1 .. $iter)) {
                 $remote->recv($data, 16);
-                system ("time ./$lib/sender $addr $threads $fps >> $lib/results/$logname 2>&1");
+                system ("time ./$lib/sender $addr $threads $fps $mode >> $lib/results/$logname 2>&1");
                 $remote->send("end") if $gen_recv;
             }
         }
@@ -141,6 +141,7 @@ sub lat_recv {
     system "$lib/latency $role >/dev/null 2>&1";
 }
 
+# TODO explain every parameter
 sub print_help {
     print "usage (benchmark):\n  ./benchmark.pl \n"
     . "\t--lib <uvgrtp|ffmpeg|gstreamer>\n"
@@ -148,6 +149,7 @@ sub print_help {
     . "\t--addr <server address>\n"
     . "\t--port <server port>\n"
     . "\t--threads <# of threads>\n"
+    . "\t--mode <strict|best-effort>\n"
     . "\t--start <start fps>\n"
     . "\t--end <end fps>\n\n";
 
@@ -172,6 +174,7 @@ GetOptions(
     "use-nc"      => \(my $nc = 0),
     "fps=s"       => \(my $fps = ""),
     "latency"     => \(my $lat = 0),
+    "mode=s"      => \(my $mode = "best-effort"),
     "help"        => \(my $help = 0)
 ) or die "failed to parse command line!\n";
 
@@ -180,6 +183,7 @@ $addr = $DEFAULT_ADDR if !$addr;
 
 print_help() if $help or !$lib;
 print_help() if ((!$start or !$end) and !$fps) and !$lat;
+print_help() if not grep /$mode/, ("strict", "best-effort");
 
 die "not implemented\n" if !grep (/$lib/, ("uvgrtp", "ffmpeg"));
 my @fps_vals = ();
@@ -204,7 +208,7 @@ if ($role eq "send") {
         lat_send($lib, $addr, $port, $role);
     } else {
         system "make $lib" . "_sender";
-        send_benchmark($lib, $addr, $port, $iter, $threads, $nc, @fps_vals);
+        send_benchmark($lib, $addr, $port, $iter, $threads, $nc, $mode, @fps_vals);
     }
 } elsif ($role eq "recv" ) {
     if ($lat) {
diff --git a/benchmarks/ffmpeg/sender.cc b/benchmarks/ffmpeg/sender.cc
index 8477d5d..e9a146a 100644
--- a/benchmarks/ffmpeg/sender.cc
+++ b/benchmarks/ffmpeg/sender.cc
@@ -21,7 +21,7 @@ extern void *get_mem(int argc, char **argv, size_t& len);
 
 std::atomic<int> nready(0);
 
-void thread_func(void *mem, size_t len, char *addr_, int thread_num, double fps)
+void thread_func(void *mem, size_t len, char *addr_, int thread_num, double fps, bool strict)
 {
     char addr[64] = { 0 };
     enum AVCodecID codec_id = AV_CODEC_ID_H265;
@@ -99,8 +99,10 @@ void thread_func(void *mem, size_t len, char *addr_, int thread_num, double fps)
             uint64_t sleep = (uint64_t)((1000 / fps) * 1000);
 
             if (diff > sleep) {
-                fprintf(stderr, "too slow (%lu vs %lu): aborting!\n", diff, sleep);
-                goto end;
+                if (strict) {
+                    fprintf(stderr, "too slow (%lu vs %lu): aborting!\n", diff, sleep);
+                    goto end;
+                }
             } else {
                 std::this_thread::sleep_for(std::chrono::microseconds(sleep - diff));
             }
@@ -129,8 +131,8 @@ end:
 
 int main(int argc, char **argv)
 {
-    if (argc != 4) {
-        fprintf(stderr, "usage: ./%s <remote address> <number of threads> <fps>\n", __FILE__);
+    if (argc != 5) {
+        fprintf(stderr, "usage: ./%s <remote address> <number of threads> <fps> <mode>\n", __FILE__);
         return -1;
     }
 
@@ -141,10 +143,11 @@ int main(int argc, char **argv)
     size_t len   = 0;
     void *mem    = get_mem(0, NULL, len);
     int nthreads = atoi(argv[2]);
+    bool strict  = !strcmp(argv[4], "strict");
     std::thread **threads = (std::thread **)malloc(sizeof(std::thread *) * nthreads);
 
     for (int i = 0; i < nthreads; ++i)
-        threads[i] = new std::thread(thread_func, mem, len, argv[1], i * 2, atof(argv[3]));
+        threads[i] = new std::thread(thread_func, mem, len, argv[1], i * 2, atof(argv[3]), strict);
 
     while (nready.load() != nthreads)
         std::this_thread::sleep_for(std::chrono::milliseconds(20));
diff --git a/benchmarks/uvgrtp/sender.cc b/benchmarks/uvgrtp/sender.cc
index 5509c67..623a6bc 100644
--- a/benchmarks/uvgrtp/sender.cc
+++ b/benchmarks/uvgrtp/sender.cc
@@ -9,7 +9,7 @@ extern void *get_mem(int argc, char **argv, size_t& len);
 
 std::atomic<int> nready(0);
 
-void thread_func(void *mem, size_t len, char *addr, int thread_num, double fps)
+void thread_func(void *mem, size_t len, char *addr, int thread_num, double fps, bool strict)
 {
     size_t bytes_sent   = 0;
     uint64_t chunk_size = 0;
@@ -50,8 +50,10 @@ void thread_func(void *mem, size_t len, char *addr, int thread_num, double fps)
             uint64_t sleep = (uint64_t)((1000 / fps) * 1000);
 
             if (diff > sleep) {
-                fprintf(stderr, "too slow (%lu vs %lu): aborting!\n", diff, sleep);
-                goto end;
+                if (strict) {
+                    fprintf(stderr, "too slow (%lu vs %lu): aborting!\n", diff, sleep);
+                    goto end;
+                }
             } else {
                 std::this_thread::sleep_for(std::chrono::microseconds(sleep - diff));
             }
@@ -76,18 +78,19 @@ end:
 
 int main(int argc, char **argv)
 {
-    if (argc != 4) {
-        fprintf(stderr, "usage: ./%s <remote address> <number of threads> <fps>\n", __FILE__);
+    if (argc != 5) {
+        fprintf(stderr, "usage: ./%s <remote address> <number of threads> <fps> <mode>\n", __FILE__);
         return -1;
     }
 
     size_t len   = 0;
     void *mem    = get_mem(0, NULL, len);
     int nthreads = atoi(argv[2]);
+    bool strict  = !strcmp(argv[4], "strict");
     std::thread **threads = (std::thread **)malloc(sizeof(std::thread *) * nthreads);
 
     for (int i = 0; i < nthreads; ++i)
-        threads[i] = new std::thread(thread_func, mem, len, argv[1], i * 2, atof(argv[3]));
+        threads[i] = new std::thread(thread_func, mem, len, argv[1], i * 2, atof(argv[3]), strict);
 
     while (nready.load() != nthreads)
         std::this_thread::sleep_for(std::chrono::milliseconds(20));