Add first CPU offset for NUMA

experiments/extra.shared.sh

 perf_stat_start() {
     local opt_args=""
     if [ "$PERF_STAT_NO_CORES" != "true" ]; then
-         opt_args="$opt_args -C 0-$((THREADS - 1))"
+         opt_args="$opt_args -C $FIRST_CPU-$((FIRST_CPU + THREADS - 1))"
     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=$!
@@ -14,7 +14,7 @@ perf_stat_stop() {
 perf_record_start() {
     local opt_args=""
     if [ "$PERF_STAT_NO_CORES" != "true" ]; then
-         opt_args="$opt_args -C 0-$((THREADS - 1))"
+         opt_args="$opt_args -C $FIRST_CPU-$((FIRST_CPU + THREADS - 1))"
     fi
 
     pushd "$DATA_OUT"

experiments/memcached.shared.sh

@@ -28,8 +28,8 @@ cleanup_memcached() {
 
 start_memcached() {
     echo "Starting memcached"
-    echo "taskset -c 0-$((THREADS - 1 - THREAD_TAIL_OFFSET)) $MEMCACHED $MEMCACHED_ARGS -u root -t $((THREADS - THREAD_TAIL_OFFSET)) -b 16384 -c 32768 -m 10240 -o hashpower=24,no_lru_maintainer,no_lru_crawler" 
-    taskset -c 0-$((THREADS - 1 - THREAD_TAIL_OFFSET)) $MEMCACHED $MEMCACHED_ARGS -u root -t $((THREADS - THREAD_TAIL_OFFSET)) -b 16384 -c 32768 -m 10240 -o hashpower=24,no_lru_maintainer,no_lru_crawler &
+    echo "taskset -c $FIRST_CPU-$((FIRST_CPU + THREADS - 1 - THREAD_TAIL_OFFSET)) $MEMCACHED $MEMCACHED_ARGS -u root -t $((THREADS - THREAD_TAIL_OFFSET)) -b 16384 -c 32768 -m 10240 -o hashpower=24,no_lru_maintainer,no_lru_crawler" 
+    taskset -c $FIRST_CPU-$((FIRST_CPU + THREADS - 1 - THREAD_TAIL_OFFSET)) $MEMCACHED $MEMCACHED_ARGS -u root -t $((THREADS - THREAD_TAIL_OFFSET)) -b 16384 -c 32768 -m 10240 -o hashpower=24,no_lru_maintainer,no_lru_crawler &
     MEMCACHED_PID=$!
     if [ "$PERF_STAT_ONLY_MEMCACHED" == "true" ]; then
         PERF_EXTRA_ARGS_ORIG="$PERF_EXTRA_ARGS"

run.sh

@@ -8,6 +8,7 @@ if [ "$UID" -ne 0 ]; then
 fi
 
 EXTRA_TUNING=()
+FIRST_CPU=0
 
 export EXP_ROOT="$PWD"
 
@@ -38,6 +39,11 @@ while [[ $# -gt 0 ]]; do
             shift
             shift
             ;;
+        --first-cpu)
+            FIRST_CPU=$2
+            shift
+            shift
+            ;;
         --fstack)
             FSTACK=true
             USERSTACK=true
@@ -122,6 +128,9 @@ if [ "${#EXTRA_TUNING[@]}" -gt 0 ]; then
         EXPERIMENT_NAME="$EXPERIMENT_NAME.$tuning"
     done
 fi
+if [ $FIRST_CPU -gt 0 ]; then
+    EXPERIMENT_NAME="$EXPERIMENT_NAME.first$FIRST_CPU"
+fi
 if [ "$FSTACK" == "true" ]; then
     EXPERIMENT_NAME="$EXPERIMENT_NAME.fstack"
 fi

tunings/irq_on_ht.sh

+if [ "$FIRST_CPU" -ne 0 ]; then
+    echo "FIRST_CPU must be 0"
+    exit 1
+fi
+
 $EXP_ROOT/irq.sh set 16 $((THREADS + 16 - 1)) 
 $EXP_ROOT/irq.sh set rx 16 $((THREADS + 16 - 1))

tunings/irq_packing.sh

-if [ "$THREADS" -ne 8 ]; then
-    echo "IRQ packing can only be done with 8 threads"
+if [ "$THREADS" -ne 8 ] || [ "$FIRST_CPU" -ne 0 ]; then
+    echo "IRQ packing can only be done with 8 threads and FIRST_CPU must be 0"
     exit 1
 fi
 

tunings/normalize_queues.sh

 ethtool -L $NET_IF rx $THREADS tx $THREADS || true
-$EXP_ROOT/irq.sh set 0 $((THREADS - 1))
-$EXP_ROOT/irq.sh set rx 0 $((THREADS - 1))
+$EXP_ROOT/irq.sh set $FIRST_CPU $((FIRST_CPU + THREADS - 1))
+$EXP_ROOT/irq.sh set rx $FIRST_CPU $((FIRST_CPU + THREADS - 1))