Have perf stat print out deltas per 5 seconds during single runs

This gives us better numbers than the hard-coded 15-second delay

experiments/extra.shared.sh

 perf_stat_start() {
     local opt_args=""
     if [ "$PERF_STAT_NO_CORES" != "true" ]; then
-         opt_args="$opt_args -C $FIRST_CPU-$((FIRST_CPU + THREADS - 1))"
+        opt_args="$opt_args -C $FIRST_CPU-$((FIRST_CPU + THREADS - 1))"
+    fi
+    if [ "$PERF_STAT_DELTA" == "true" ]; then
+        opt_args="$opt_args -I 5000"
     fi
     $PERF stat $PERF_EXTRA_ARGS $opt_args -e LLC-load-misses:u,LLC-load-misses:k,LLC-loads:u,LLC-loads:k,LLC-store-misses:u,LLC-store-misses:k,LLC-stores:u,LLC-stores:k,task-clock,cycles:u,cycles:k,instructions:u,instructions:k -o "$DATA_OUT/stat_$1$EXP_AUX_SUFFIX.txt" &
     PERF_STAT_PID=$!
@@ -14,7 +17,7 @@ perf_stat_stop() {
 perf_record_start() {
     local opt_args=""
     if [ "$PERF_STAT_NO_CORES" != "true" ]; then
-         opt_args="$opt_args -C $FIRST_CPU-$((FIRST_CPU + THREADS - 1))"
+        opt_args="$opt_args -C $FIRST_CPU-$((FIRST_CPU + THREADS - 1))"
     fi
 
     pushd "$DATA_OUT"

experiments/memcached.shared.sh

@@ -76,7 +76,6 @@ memcached_pre_start() {
 
     if [ "$FSTACK" == "true" ]; then
         MEMCACHED="$MEMCACHED_FSTACK"
-        PERF_EXTRA_ARGS="-D 15000" # Account for S-Stack connection establishment delay
     elif [ "$CALADAN" == "true" ]; then
         MEMCACHED="$MEMCACHED_CALADAN"
         MEMCACHED_ARGS="$MEMCACHED_ARGS $CALADAN_CONFIG -p 11211"

experiments/memcached_single_run.sh

@@ -3,6 +3,8 @@ source "$EXP_ROOT/experiments/memcached.shared.sh"
 pre_start() {
     memcached_pre_start
     EXPERIMENT_NAME_EXT="t$THREADS"
+    # 5-second interval delta for perf stat, so that we can manually subtract startup phases
+    PERF_STAT_DELTA=true
 
     while [ ! -z "$1" ]; do
         case "$1" in