[dpdk-dev] [PATCH v2 1/2] testpmd: add forwarding mode to simulate a noisy neighbour

[Jens Freimann]

This adds a new forwarding mode to testpmd to simulate
more realistic behavior of a guest machine engaged in receiving
and sending packets performing Virtual Network Function (VNF).

The goal is to enable a simple way of measuring performance impact on
cache and memory footprint utilization from various VNF co-located on
the same host machine. For this it does:

* Buffer packets in a FIFO:

Create a fifo to buffer received packets. Once it flows over put
those packets into the actual tx queue. The fifo is created per tx
queue and its size can be set with the --buffersize-before-sending
commandline parameter.

A second commandline parameter is used to set a timeout in
milliseconds after which the fifo is flushed.

--buffersize-before-sending [packet numbers]
Keep the mbuf in a FIFO and forward the over flooding packets from the
FIFO. This queue is per TX-queue (after all other packet processing).

--flush-timeout [delay]
Flush the packet queue if no packets have been seen during
[delay]. As long as packets are seen, the timer is reset.

Add several options to simulate route lookups (memory reads) in tables
that can be quite large, as well as route hit statistics update.
These options simulates the while stack traversal and
will trash the cache. Memory access is random.

* simulate route lookups:

Allocate a buffer and perform reads and writes on it as specified by
commandline options:

--memory-footprint [size]
Size of the VNF internal memory (MB), in which the random
read/write will be done, allocated by rte_malloc (hugepages).

--nb-rnd-write [num]
Number of random writes in memory per packet should be
performed, simulating hit-flags update. 64 bits per write,
all write in different cache lines.

--nb-rnd-read [num]
Number of random reads in memory per packet should be
performed, simulating FIB/table lookups. 64 bits per read,
all write in different cache lines.

--nb-rnd-read-write [num]
Number of random reads and writes in memory per packet should
be performed, simulating stats update. 64 bits per read-write, all
reads and writes in different cache lines.

Signed-off-by: Jens Freimann <jfreimann at redhat.com>
---
 app/test-pmd/Makefile     |   1 +
 app/test-pmd/noisy_vnf.c  | 184 ++++++++++++++++++++++++++++++++++++++++++++++
 app/test-pmd/noisy_vnf.h  |  42 +++++++++++
 app/test-pmd/parameters.c |  55 +++++++++++++-
 app/test-pmd/testpmd.c    |  68 +++++++++++++++++
 app/test-pmd/testpmd.h    |  17 +++++
 6 files changed, 366 insertions(+), 1 deletion(-)
 create mode 100644 app/test-pmd/noisy_vnf.c
 create mode 100644 app/test-pmd/noisy_vnf.h

diff --git a/app/test-pmd/Makefile b/app/test-pmd/Makefile
index 60ae9b9c1..2dc07ab16 100644
--- a/app/test-pmd/Makefile
+++ b/app/test-pmd/Makefile
@@ -32,6 +32,7 @@ SRCS-y += rxonly.c
 SRCS-y += txonly.c
 SRCS-y += csumonly.c
 SRCS-y += icmpecho.c
+SRCS-y += noisy_vnf.c
 SRCS-$(CONFIG_RTE_LIBRTE_IEEE1588) += ieee1588fwd.c
 
 ifeq ($(CONFIG_RTE_LIBRTE_PMD_SOFTNIC)$(CONFIG_RTE_LIBRTE_SCHED),yy)
diff --git a/app/test-pmd/noisy_vnf.c b/app/test-pmd/noisy_vnf.c
new file mode 100644
index 000000000..69ba29859
--- /dev/null
+++ b/app/test-pmd/noisy_vnf.c
@@ -0,0 +1,184 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Red Hat Corp.
+ */
+
+#include <stdarg.h>
+#include <stdio.h>
+#include <string.h>
+#include <errno.h>
+#include <stdint.h>
+#include <unistd.h>
+#include <inttypes.h>
+
+#include <sys/queue.h>
+#include <sys/stat.h>
+
+#include <rte_common.h>
+#include <rte_log.h>
+#include <rte_debug.h>
+#include <rte_cycles.h>
+#include <rte_memory.h>
+#include <rte_launch.h>
+#include <rte_eal.h>
+#include <rte_per_lcore.h>
+#include <rte_lcore.h>
+#include <rte_memcpy.h>
+#include <rte_mempool.h>
+#include <rte_mbuf.h>
+#include <rte_ethdev.h>
+#include <rte_flow.h>
+#include <rte_malloc.h>
+
+#include "testpmd.h"
+#include "noisy_vnf.h"
+
+static inline void
+do_write(char *vnf_mem)
+{
+	uint64_t i = rte_rand();
+	uint64_t w = rte_rand();
+
+	vnf_mem[i % ((vnf_memory_footprint * 1024 * 1024) /
+			RTE_CACHE_LINE_SIZE)] = w;
+}
+
+	static inline void
+do_read(char *vnf_mem)
+{
+	uint64_t i = rte_rand();
+	uint64_t r;
+
+	r = vnf_mem[i % ((vnf_memory_footprint * 1024 * 1024) /
+			RTE_CACHE_LINE_SIZE)];
+	r++;
+}
+
+	static inline void
+do_rw(char *vnf_mem)
+{
+	do_read(vnf_mem);
+	do_write(vnf_mem);
+}
+
+/*
+ * Simulate route lookups as defined by commandline parameters
+ */
+	static void
+sim_memory_lookups(struct noisy_config *ncf, uint16_t nb_pkts)
+{
+	uint16_t i, j;
+
+	for (i = 0; i < nb_pkts; i++) {
+		for (j = 0; j < nb_rnd_write; j++)
+			do_write(ncf->vnf_mem);
+		for (j = 0; j < nb_rnd_read; j++)
+			do_read(ncf->vnf_mem);
+		for (j = 0; j < nb_rnd_read_write; j++)
+			do_rw(ncf->vnf_mem);
+	}
+}
+
+/*
+ * Forwarding of packets in I/O mode.
+ * Forward packets "as-is".
+ * This is the fastest possible forwarding operation, as it does not access
+ * to packets data.
+ */
+static void
+pkt_burst_noisy_vnf(struct fwd_stream *fs)
+{
+	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
+	uint16_t nb_rx;
+	uint16_t nb_tx = 0;
+	uint32_t retry;
+	const uint64_t freq_khz = rte_get_timer_hz() / 1000;
+	struct noisy_config *ncf = &noisy_cfg[fs->tx_queue];
+	struct rte_mbuf *tmp_pkts[MAX_PKT_BURST];
+	uint16_t nb_enqd;
+	uint16_t nb_deqd = 0;
+	uint64_t delta_ms;
+	uint64_t now;
+
+	/*
+	 * Receive a burst of packets and forward them.
+	 */
+	nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue,
+			pkts_burst, nb_pkt_per_burst);
+	if (unlikely(nb_rx == 0))
+		return;
+	fs->rx_packets += nb_rx;
+
+	if (bsize_before_send > 0) {
+		if (rte_ring_free_count(ncf->f) >= nb_rx) {
+			/* enqueue into fifo */
+			nb_enqd = fifo_put(ncf->f, pkts_burst, nb_rx);
+			if (nb_enqd < nb_rx)
+				nb_rx = nb_enqd;
+		} else {
+			/* fifo is full, dequeue first */
+			nb_deqd = fifo_get(ncf->f, tmp_pkts, nb_rx);
+			/* enqueue into fifo */
+			nb_enqd = fifo_put(ncf->f, pkts_burst, nb_deqd);
+			sim_memory_lookups(ncf, nb_rx);
+			if (nb_enqd < nb_rx)
+				nb_rx = nb_enqd;
+			if (nb_deqd > 0)
+				nb_tx = rte_eth_tx_burst(fs->tx_port,
+						fs->tx_queue, tmp_pkts,
+						nb_deqd);
+		}
+	} else {
+		sim_memory_lookups(ncf, nb_rx);
+		nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
+				pkts_burst, nb_rx);
+	}
+
+	/*
+	 * TX burst queue drain
+	 */
+	if (ncf->prev_time == 0)
+		now = ncf->prev_time = rte_get_timer_cycles();
+	else
+		now = rte_get_timer_cycles();
+	delta_ms = (now - ncf->prev_time) / freq_khz;
+	if (unlikely(delta_ms >= flush_timer) && flush_timer > 0 &&
+		     (nb_tx == 0)) {
+		while (fifo_count(ncf->f) > 0) {
+			nb_deqd = fifo_get(ncf->f, tmp_pkts, nb_rx);
+			nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
+						 tmp_pkts, nb_deqd);
+			if (rte_ring_empty(ncf->f))
+				break;
+		}
+		ncf->prev_time = now;
+	}
+	if (nb_tx < nb_rx && fs->retry_enabled)
+		*pkts_burst = *tmp_pkts;
+
+	/*
+	 * Retry if necessary
+	 */
+	if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
+		retry = 0;
+		while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
+			rte_delay_us(burst_tx_delay_time);
+			sim_memory_lookups(ncf, nb_rx);
+			nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
+					&pkts_burst[nb_tx], nb_rx - nb_tx);
+		}
+	}
+	fs->tx_packets += nb_tx;
+	if (unlikely(nb_tx < nb_rx)) {
+		fs->fwd_dropped += (nb_rx - nb_tx);
+		do {
+			rte_pktmbuf_free(pkts_burst[nb_tx]);
+		} while (++nb_tx < nb_rx);
+	}
+}
+
+struct fwd_engine noisy_vnf_engine = {
+	.fwd_mode_name  = "noisy",
+	.port_fwd_begin = NULL,
+	.port_fwd_end   = NULL,
+	.packet_fwd     = pkt_burst_noisy_vnf,
+};
diff --git a/app/test-pmd/noisy_vnf.h b/app/test-pmd/noisy_vnf.h
new file mode 100644
index 000000000..7f73b4301
--- /dev/null
+++ b/app/test-pmd/noisy_vnf.h
@@ -0,0 +1,42 @@
+#ifndef __NOISY_VNF_H
+#define __NOISY_VNF_H
+#include <stdint.h>
+#include <rte_mbuf.h>
+#include <rte_ring.h>
+#include "testpmd.h"
+
+#define FIFO_COUNT_MAX 1024
+/**
+ * Add elements to fifo. Return number of written elements
+ */
+static inline unsigned int
+fifo_put(struct rte_ring *r, struct rte_mbuf **data, unsigned num)
+{
+
+	return rte_ring_enqueue_burst(r, (void **)data, num, NULL);
+}
+
+/**
+ * Get elements from fifo. Return number of read elements
+ */
+static inline unsigned int
+fifo_get(struct rte_ring *r, struct rte_mbuf **data, unsigned num)
+{
+	return rte_ring_dequeue_burst(r, (void **) data, num, NULL);
+}
+
+static inline unsigned int
+fifo_count(struct rte_ring *r)
+{
+	return rte_ring_count(r);
+}
+
+static inline int
+fifo_full(struct rte_ring *r)
+{
+	return rte_ring_full(r);
+}
+
+struct rte_ring *noisy_init(uint32_t qi, uint32_t pi);
+
+#endif
diff --git a/app/test-pmd/parameters.c b/app/test-pmd/parameters.c
index 2192bdcdf..0583e594f 100644
--- a/app/test-pmd/parameters.c
+++ b/app/test-pmd/parameters.c
@@ -621,6 +621,12 @@ launch_args_parse(int argc, char** argv)
 		{ "print-event",		1, 0, 0 },
 		{ "mask-event",			1, 0, 0 },
 		{ "tx-offloads",		1, 0, 0 },
+		{ "buffersize-before-sending",  1, 0, 0 },
+		{ "flush-timeout",              1, 0, 0 },
+		{ "memory-footprint",           1, 0, 0 },
+		{ "nb-rnd-write",               1, 0, 0 },
+		{ "nb-rnd-read",                1, 0, 0 },
+		{ "nb-rnd-read-write",          1, 0, 0 },
 		{ 0, 0, 0, 0 },
 	};
 
@@ -1101,7 +1107,54 @@ launch_args_parse(int argc, char** argv)
 					rte_exit(EXIT_FAILURE,
 						 "invalid mask-event argument\n");
 				}
-
+			if (!strcmp(lgopts[opt_idx].name, "buffersize-before-sending")) {
+				n = atoi(optarg);
+				if (n > 0)
+					bsize_before_send = (uint16_t) n;
+				else
+					rte_exit(EXIT_FAILURE,
+						 "buffersize-before-sending must be > 0\n");
+			}
+			if (!strcmp(lgopts[opt_idx].name, "flush-timeout")) {
+				n = atoi(optarg);
+				if (n >= 0)
+					flush_timer = (uint16_t) n;
+				else
+					rte_exit(EXIT_FAILURE,
+						 "flush-timeout must be > 0\n");
+			}
+			if (!strcmp(lgopts[opt_idx].name, "memory-footprint")) {
+				n = atoi(optarg);
+				if (n > 0)
+					vnf_memory_footprint = (uint16_t) n;
+				else
+					rte_exit(EXIT_FAILURE,
+						 "memory-footprint must be > 0\n");
+			}
+			if (!strcmp(lgopts[opt_idx].name, "nb-rnd-write")) {
+				n = atoi(optarg);
+				if (n > 0)
+					nb_rnd_write = (uint16_t) n;
+				else
+					rte_exit(EXIT_FAILURE,
+						 "nb-rnd-write must be > 0\n");
+			}
+			if (!strcmp(lgopts[opt_idx].name, "nb-rnd-read")) {
+				n = atoi(optarg);
+				if (n > 0)
+					nb_rnd_read = (uint16_t) n;
+				else
+					rte_exit(EXIT_FAILURE,
+						 "nb-rnd-read must be > 0\n");
+			}
+			if (!strcmp(lgopts[opt_idx].name, "nb-rnd-read-write")) {
+				n = atoi(optarg);
+				if (n > 0)
+					nb_rnd_read_write = (uint16_t) n;
+				else
+					rte_exit(EXIT_FAILURE,
+						 "nb-rnd-read-write must be > 0\n");
+			}
 			break;
 		case 'h':
 			usage(argv[0]);
diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
index 4c0e2586c..29ab4a262 100644
--- a/app/test-pmd/testpmd.c
+++ b/app/test-pmd/testpmd.c
@@ -60,6 +60,7 @@
 #include <rte_latencystats.h>
 #endif
 
+#include "noisy_vnf.h"
 #include "testpmd.h"
 
 uint16_t verbose_level = 0; /**< Silent by default. */
@@ -154,6 +155,7 @@ struct fwd_engine * fwd_engines[] = {
 	&tx_only_engine,
 	&csum_fwd_engine,
 	&icmp_echo_engine,
+	&noisy_vnf_engine,
 #if defined RTE_LIBRTE_PMD_SOFTNIC && defined RTE_LIBRTE_SCHED
 	&softnic_tm_engine,
 	&softnic_tm_bypass_engine,
@@ -249,6 +251,40 @@ int16_t tx_free_thresh = RTE_PMD_PARAM_UNSET;
  */
 int16_t tx_rs_thresh = RTE_PMD_PARAM_UNSET;
 
+/*
+ * Configurable value of buffered packed before sending.
+ */
+uint16_t bsize_before_send = 0;
+
+/*
+ * Configurable value of packet buffer timeout.
+ */
+uint16_t flush_timer = 0;
+
+/*
+ * Configurable value for size of VNF internal memory area
+ * used for simulating noisy neighbour behaviour
+ */
+uint64_t vnf_memory_footprint = 0;
+
+/*
+ * Configurable value of number of random writes done in
+ * VNF simulation memory area.
+ */
+uint64_t nb_rnd_write = 0;
+
+/*
+ * Configurable value of number of random reads done in
+ * VNF simulation memory area.
+ */
+uint64_t nb_rnd_read = 0;
+
+/*
+ * Configurable value of number of random reads/wirtes done in
+ * VNF simulation memory area.
+ */
+uint64_t nb_rnd_read_write = 0;
+
 /*
  * Receive Side Scaling (RSS) configuration.
  */
@@ -401,6 +437,24 @@ static int all_ports_started(void);
 struct gso_status gso_ports[RTE_MAX_ETHPORTS];
 uint16_t gso_max_segment_size = ETHER_MAX_LEN - ETHER_CRC_LEN;
 
+#define NOISY_STRSIZE 256
+#define NOISY_RING "noisy_ring_%d:%d\n"
+struct rte_ring * noisy_init(uint32_t qi, uint32_t pi)
+{
+	struct noisy_config *n = &noisy_cfg[qi];
+	char name[NOISY_STRSIZE];
+
+	snprintf(name, NOISY_STRSIZE, NOISY_RING, pi, qi);
+	n->f = rte_ring_create(name, bsize_before_send, rte_socket_id(), 0);
+	n->vnf_mem = (char *) rte_zmalloc("vnf sim memory",
+			 vnf_memory_footprint * 1024 * 1024,
+			 RTE_CACHE_LINE_SIZE);
+	if (n->vnf_mem == NULL)
+		printf("allocating vnf memory failed\n");
+
+	return n->f;
+}
+
 /*
  * Helper function to check if socket is already discovered.
  * If yes, return positive value. If not, return zero.
@@ -1584,6 +1638,15 @@ start_port(portid_t pid)
 				return -1;
 			}
 		}
+		noisy_cfg = (struct noisy_config *) rte_zmalloc(
+				"testpmd noisy fifo and timers",
+				nb_txq * sizeof(struct noisy_config),
+				RTE_CACHE_LINE_SIZE);
+		if (noisy_cfg == NULL) {
+			rte_exit(EXIT_FAILURE,
+					"rte_zmalloc(%d) struct noisy_config) failed\n",
+					(int)(nb_txq * sizeof(struct noisy_config)));
+		}
 		if (port->need_reconfig_queues > 0) {
 			port->need_reconfig_queues = 0;
 			port->tx_conf.txq_flags = ETH_TXQ_FLAGS_IGNORE;
@@ -1591,6 +1654,9 @@ start_port(portid_t pid)
 			port->tx_conf.offloads = port->dev_conf.txmode.offloads;
 			/* setup tx queues */
 			for (qi = 0; qi < nb_txq; qi++) {
+				if (!noisy_init(qi, pi) && bsize_before_send > 0)
+					rte_exit(EXIT_FAILURE, "%s\n",
+						 rte_strerror(rte_errno));
 				if ((numa_support) &&
 					(txring_numa[pi] != NUMA_NO_CONFIG))
 					diag = rte_eth_tx_queue_setup(pi, qi,
@@ -1755,6 +1821,8 @@ stop_port(portid_t pid)
 			RTE_PORT_HANDLING, RTE_PORT_STOPPED) == 0)
 			printf("Port %d can not be set into stopped\n", pi);
 		need_check_link_status = 1;
+
+		rte_free(noisy_cfg);
 	}
 	if (need_check_link_status && !no_link_check)
 		check_all_ports_link_status(RTE_PORT_ALL);
diff --git a/app/test-pmd/testpmd.h b/app/test-pmd/testpmd.h
index 153abea05..12d601e0d 100644
--- a/app/test-pmd/testpmd.h
+++ b/app/test-pmd/testpmd.h
@@ -9,6 +9,7 @@
 #include <rte_bus_pci.h>
 #include <rte_gro.h>
 #include <rte_gso.h>
+#include "noisy_vnf.h"
 
 #define RTE_PORT_ALL            (~(portid_t)0x0)
 
@@ -109,6 +110,14 @@ struct fwd_stream {
 #endif
 };
 
+struct noisy_config {
+	struct rte_ring *f;
+	uint64_t prev_time;
+	char *vnf_mem;
+};
+struct noisy_config *noisy_cfg;
+
+
 /** Descriptor for a single flow. */
 struct port_flow {
 	size_t size; /**< Allocated space including data[]. */
@@ -251,6 +260,7 @@ extern struct fwd_engine rx_only_engine;
 extern struct fwd_engine tx_only_engine;
 extern struct fwd_engine csum_fwd_engine;
 extern struct fwd_engine icmp_echo_engine;
+extern struct fwd_engine noisy_vnf_engine;
 #ifdef TM_MODE
 extern struct fwd_engine softnic_tm_engine;
 extern struct fwd_engine softnic_tm_bypass_engine;
@@ -382,6 +392,13 @@ extern int8_t rx_drop_en;
 extern int16_t tx_free_thresh;
 extern int16_t tx_rs_thresh;
 
+extern uint16_t bsize_before_send;
+extern uint16_t flush_timer;
+extern uint64_t vnf_memory_footprint;
+extern uint64_t nb_rnd_write;
+extern uint64_t nb_rnd_read;
+extern uint64_t nb_rnd_read_write;
+
 extern uint8_t dcb_config;
 extern uint8_t dcb_test;
 
-- 
2.14.3