diff --git a/src/vnet/util/refcount.h b/src/vnet/util/refcount.h
new file mode 100644
index 0000000..ea92148
--- /dev/null
+++ b/src/vnet/util/refcount.h
@@ -0,0 +1,95 @@
+/*
+ * Copyright (c) 2016 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/*
+ * vlib provides lock-free counters but those
+ * - Have 16bits per-CPU counter, which may overflow.
+ * - Would only increment.
+ *
+ * This is very similar to vlib counters, but may be used to count reference.
+ * Such a counter includes an arbitrary number of counters. Each counter
+ * is identified by its index. This is used to aggregate per-cpu memory.
+ *
+ * Warning:
+ *   This reference counter is lock-free but is not race-condition free.
+ *   The counting result is approximate and another mechanism needs to be used
+ *   in order to ensure that an object may be freed.
+ *
+ */
+
+#include <vnet/vnet.h>
+
+/*
+ * Reference counting
+ * A specific reference counter is used. The design is quite
+ * similar to vlib counters but:
+ *   - It is possible to decrease the value
+ *   - Summing will not zero the per-thread counters
+ *   - Only the thread can reallocate its own counters vector (to avoid concurrency issues)
+*/
+typedef struct {
+  u32 *counters;
+  volatile u32 *counter_lock;
+  CLIB_CACHE_LINE_ALIGN_MARK(o);
+} vlib_refcount_per_cpu_t;
+
+typedef struct {
+  vlib_refcount_per_cpu_t *per_cpu;
+} vlib_refcount_t;
+
+static_always_inline
+void vlib_refcount_lock (volatile u32 *counter_lock)
+{
+  while (__sync_lock_test_and_set (counter_lock, 1))
+    ;
+}
+
+static_always_inline
+void vlib_refcount_unlock (volatile u32 *counter_lock)
+{
+  *counter_lock = 0;
+}
+
+void __vlib_refcount_resize(vlib_refcount_per_cpu_t *per_cpu, u32 size);
+
+static_always_inline
+void vlib_refcount_add(vlib_refcount_t *r, u32 thread_index, u32 counter_index, i32 v)
+{
+  vlib_refcount_per_cpu_t *per_cpu = &r->per_cpu[thread_index];
+  if (PREDICT_FALSE(counter_index >= vec_len(per_cpu->counters)))
+    __vlib_refcount_resize(per_cpu, clib_max(counter_index + 16,(vec_len(per_cpu->counters)) * 2));
+
+  per_cpu->counters[counter_index] += v;
+}
+
+u64 vlib_refcount_get(vlib_refcount_t *r, u32 index);
+
+static_always_inline
+void vlib_refcount_init(vlib_refcount_t *r)
+{
+  vlib_thread_main_t *tm = vlib_get_thread_main ();
+  u32 thread_index;
+  r->per_cpu = 0;
+  vec_validate (r->per_cpu, tm->n_vlib_mains - 1);
+
+  for (thread_index = 0; thread_index < tm->n_vlib_mains; thread_index++)
+    {
+      r->per_cpu[thread_index].counter_lock =
+	  clib_mem_alloc_aligned(CLIB_CACHE_LINE_BYTES,CLIB_CACHE_LINE_BYTES);
+      r->per_cpu[thread_index].counter_lock[0] = 0;
+    }
+}
+
+