diff --git a/src/dp_port.c b/src/dp_port.c
index 6717e193f..8a8e12501 100644
--- a/src/dp_port.c
+++ b/src/dp_port.c
@@ -99,6 +99,8 @@ static int dp_port_init_ethdev(struct dp_port *port, struct rte_eth_dev_info *de
 
 	/* Default config */
 	port_conf.txmode.offloads &= dev_info->tx_offload_capa;
+	if (dev_info->rx_offload_capa & RTE_ETH_RX_OFFLOAD_TIMESTAMP)
+		port_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_TIMESTAMP;
 
 	if (dp_conf_get_nic_type() == DP_CONF_NIC_TYPE_TAP || dp_conf_is_multiport_eswitch())
 		nr_hairpin_queues = 0;
diff --git a/src/dpdk_layer.c b/src/dpdk_layer.c
index 129e38cc1..4dcc747c1 100644
--- a/src/dpdk_layer.c
+++ b/src/dpdk_layer.c
@@ -146,8 +146,10 @@ static int graph_main_loop(__rte_unused void *arg)
 		nanosleep(&standby_sleep, NULL);
 	}
 
-	DPS_LOG_INFO("Starting packet processing");
-	rx_node_start_processing();
+	if (!force_quit) {
+		DPS_LOG_INFO("Starting packet processing");
+		rx_node_start_processing();
+	}
 
 	while (!force_quit)
 		rte_graph_walk(graph);
diff --git a/src/nodes/rx_node.c b/src/nodes/rx_node.c
index 0212f2b01..3ab773203 100644
--- a/src/nodes/rx_node.c
+++ b/src/nodes/rx_node.c
@@ -21,19 +21,56 @@ DP_NODE_REGISTER_SOURCE(RX, rx, NEXT_NODES);
 struct rx_node_ctx {
 	const struct dp_port *port;
 	uint16_t	queue_id;
+	bool		flush_old_packets;
 };
 static_assert(sizeof(struct rx_node_ctx) <= RTE_NODE_CTX_SZ,
 			  "Rx node context will not fit into the node");
 
 // also some way to map ports to nodes is needed
 static rte_node_t rx_node_ids[DP_MAX_PORTS];
+static struct rx_node_ctx *rx_node_ctxs[DP_MAX_PORTS];
 
 // dpservice starts in "standby mode" (no processing of traffic)
 static volatile bool standing_by = true;
+static int rx_timestamp_offset;
+static uint64_t flush_timestamp;
 
 
+static uint64_t get_current_timestamp(void)
+{
+	const struct dp_ports *ports;
+	uint64_t timestamp = 0;
+	int ret;
+
+	// is timestamping of packets even supported?
+	if (rx_timestamp_offset <= 0)
+		return 0;
+
+	// PF0/PF1 are in isolated mode, which prevents rte_eth_read_clock() from working
+	// thus find the first VF that is up (timestamp is the same for all ports on the NIC)
+	ports = dp_get_ports();
+	DP_FOREACH_PORT(ports, port) {
+		if (!port->allocated)
+			continue;
+		ret = rte_eth_read_clock(port->port_id, &timestamp);
+		if (!DP_FAILED(ret))
+			break;
+	}
+	return timestamp;
+}
+
 void rx_node_start_processing(void)
 {
+	// even though processing was stopped, buffers still contain old packets
+	// to flush them, need to know which ones are old, need current timestamp
+	flush_timestamp = get_current_timestamp();
+	if (flush_timestamp > 0) {
+		// notify all Rx nodes that they need to flush
+		for (size_t i = 0; i < RTE_DIM(rx_node_ctxs); ++i)
+			if (rx_node_ctxs[i])
+				rx_node_ctxs[i]->flush_old_packets = true;
+	}
+
 	standing_by = false;
 }
 
@@ -65,6 +102,13 @@ static int rx_node_init(const struct rte_graph *graph, struct rte_node *node)
 	uint16_t port_id;
 	const struct dp_port *port;
 
+	if (rx_timestamp_offset == 0) {
+		if (DP_FAILED(rte_mbuf_dyn_rx_timestamp_register(&rx_timestamp_offset, NULL))) {
+			DPS_LOG_ERR("Cannot register Rx timestamp field", DP_LOG_RET(rte_errno));
+			return DP_ERROR;
+		}
+	}
+
 	// Find this node's dedicated port to be used in processing
 	for (port_id = 0; port_id < RTE_DIM(rx_node_ids); ++port_id)
 		if (rx_node_ids[port_id] == node->id)
@@ -84,10 +128,26 @@ static int rx_node_init(const struct rte_graph *graph, struct rte_node *node)
 	// save dp_port to this node's context for accessing its id and the status of allocation
 	ctx->port = port;
 	ctx->queue_id = graph->id;
+	rx_node_ctxs[port_id] = ctx;
 	DPNODE_LOG_INFO(node, "Initialized", DP_LOG_PORTID(ctx->port->port_id), DP_LOG_QUEUEID(ctx->queue_id));
 	return DP_OK;
 }
 
+
+static uint16_t rx_find_old_packets(void **objs, uint16_t n_pkts, uint64_t timestamp)
+{
+	rte_mbuf_timestamp_t *pkt_timestamp;
+	uint16_t old = 0;
+
+	for (uint16_t i = 0; i < n_pkts; ++i) {
+		pkt_timestamp = RTE_MBUF_DYNFIELD(objs[i], rx_timestamp_offset, rte_mbuf_timestamp_t *);
+		if (*pkt_timestamp >= timestamp)
+			break;
+		old++;
+	}
+	return old;
+}
+
 static uint16_t rx_node_process(struct rte_graph *graph,
 								struct rte_node *node,
 								void **objs,
@@ -95,6 +155,7 @@ static uint16_t rx_node_process(struct rte_graph *graph,
 {
 	struct rx_node_ctx *ctx = (struct rx_node_ctx *)node->ctx;
 	uint16_t n_pkts;
+	uint16_t old;
 
 	RTE_SET_USED(cnt);  // this is a source node, input data is not present yet
 
@@ -104,13 +165,25 @@ static uint16_t rx_node_process(struct rte_graph *graph,
 	if (unlikely(standing_by))
 		return 0;
 
-	n_pkts = rte_eth_rx_burst(ctx->port->port_id,
-							  ctx->queue_id,
-							  (struct rte_mbuf **)objs,
-							  RTE_GRAPH_BURST_SIZE);
+	n_pkts = rte_eth_rx_burst(ctx->port->port_id, ctx->queue_id, (struct rte_mbuf **)objs, RTE_GRAPH_BURST_SIZE);
 	if (unlikely(!n_pkts))
 		return 0;
 
+	if (unlikely(ctx->flush_old_packets)) {
+		DPS_LOG_INFO("Flushing old packets", DP_LOG_PORT(ctx->port));
+		old = rx_find_old_packets(objs, n_pkts, flush_timestamp);
+		if (old > 0) {
+			rte_pktmbuf_free_bulk((struct rte_mbuf **)objs, old);
+			objs += old;
+			n_pkts -= old;
+			DPS_LOG_INFO("Flushed old packets", DP_LOG_VALUE(old), DP_LOG_PORT(ctx->port));
+			// if all packets were old, continue flushing
+			if (old == n_pkts)
+				return 0;
+		}
+		ctx->flush_old_packets = false;
+	}
+
 	node->idx = n_pkts;
 
 	// Rx node only ever leads to CLS node (can move all packets at once)