DPDK patches and discussions
 help / color / mirror / Atom feed
From: Daniel Mrzyglod <danielx.t.mrzyglod@intel.com>
To: dev@dpdk.org
Subject: [dpdk-dev] [PATCH v4 1/2] This patch add support of mode 5 to link bonding pmd
Date: Thu, 27 Nov 2014 16:33:40 +0000	[thread overview]
Message-ID: <1417106021-22842-2-git-send-email-danielx.t.mrzyglod@intel.com> (raw)
In-Reply-To: <1417106021-22842-1-git-send-email-danielx.t.mrzyglod@intel.com>

Add support for mode 5 (Transmit load balancing) into pmd driver

This patch add support for Adaptive transmit load balancing (mode 5) to the
librte_pmd_bond library. This mode provides an adaptive transmit load 
balancing. It dynamically changes the transmitting slave, according to the 
computed load. 

Further details are described here:
https://www.kernel.org/doc/Documentation/networking/bonding.txt
In implementation callback is used for sorting slave order - providing 
statistics for burst function about slave bandwith usage  and sort 
interfaces due to usage.

Difference in this implementation vs Linux implementation:
- We Are trying send all pkts – If one interface hasn’t send packets we are 
trying to send rest of packets by other slaves sorted previously by callback 
function.

Some implementation details:
- Every 100ms is taken obytes statistics from every slave.
- Every 10 ms the slaves in  table are sorted and updated by callback - 
bandwidth and successfully transmitted bytes from previous iteration which 
happens every 100 ms
- There is callback function which updates this statistics for transparency and
for rather intensive computation involved in this mode.

Signed-off-by: Daniel Mrzyglod <danielx.t.mrzyglod@intel.com>
---
 lib/librte_pmd_bond/rte_eth_bond.h         |   6 ++
 lib/librte_pmd_bond/rte_eth_bond_args.c    |   1 +
 lib/librte_pmd_bond/rte_eth_bond_pmd.c     | 160 ++++++++++++++++++++++++++++-
 lib/librte_pmd_bond/rte_eth_bond_private.h |   2 +-
 4 files changed, 166 insertions(+), 3 deletions(-)

diff --git a/lib/librte_pmd_bond/rte_eth_bond.h b/lib/librte_pmd_bond/rte_eth_bond.h
index 085500b..3831f56 100644
--- a/lib/librte_pmd_bond/rte_eth_bond.h
+++ b/lib/librte_pmd_bond/rte_eth_bond.h
@@ -77,6 +77,12 @@ extern "C" {
  * In this mode all transmitted packets will be transmitted on all available
  * active slaves of the bonded. */
 #endif
+#define BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING	(5)
+/**< Adaptive TLB (Mode 5)
+ * This mode provides an adaptive transmit load balancing. It dynamically
+ * changes the transmitting slave, according to the computed load. Statistics
+ * are collected in 100ms intervals and scheduled every 10ms */
+
 /* Balance Mode Transmit Policies */
 #define BALANCE_XMIT_POLICY_LAYER2		(0)
 /**< Layer 2 (Ethernet MAC) */
diff --git a/lib/librte_pmd_bond/rte_eth_bond_args.c b/lib/librte_pmd_bond/rte_eth_bond_args.c
index d8ce681..2675cf6 100644
--- a/lib/librte_pmd_bond/rte_eth_bond_args.c
+++ b/lib/librte_pmd_bond/rte_eth_bond_args.c
@@ -173,6 +173,7 @@ bond_ethdev_parse_slave_mode_kvarg(const char *key __rte_unused,
 #ifdef RTE_MBUF_REFCNT
 	case BONDING_MODE_BROADCAST:
 #endif
+	case BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING:
 		return 0;
 	default:
 		RTE_BOND_LOG(ERR, "Invalid slave mode value (%s) specified", value);
diff --git a/lib/librte_pmd_bond/rte_eth_bond_pmd.c b/lib/librte_pmd_bond/rte_eth_bond_pmd.c
index cf2fbab..7a5dae6 100644
--- a/lib/librte_pmd_bond/rte_eth_bond_pmd.c
+++ b/lib/librte_pmd_bond/rte_eth_bond_pmd.c
@@ -30,7 +30,7 @@
  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
-
+#include <stdlib.h>
 #include <rte_mbuf.h>
 #include <rte_malloc.h>
 #include <rte_ethdev.h>
@@ -41,10 +41,15 @@
 #include <rte_kvargs.h>
 #include <rte_dev.h>
 #include <rte_alarm.h>
+#include <rte_cycles.h>
 
 #include "rte_eth_bond.h"
 #include "rte_eth_bond_private.h"
 
+#define REORDER_PERIOD_MS 10
+/* Table for statistics in mode 5 TLB */
+static uint64_t tlb_last_obytets[RTE_MAX_ETHPORTS];
+
 static uint16_t
 bond_ethdev_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 {
@@ -288,6 +293,144 @@ xmit_slave_hash(const struct rte_mbuf *buf, uint8_t slave_count, uint8_t policy)
 	return hash % slave_count;
 }
 
+struct bwg_slave {
+	uint64_t bwg_left_int;
+	uint64_t bwg_left_remainder;
+	uint8_t slave;
+};
+
+static int
+bandwidth_cmp(const void *a, const void *b)
+{
+	const struct bwg_slave *bwg_a = a;
+	const struct bwg_slave *bwg_b = b;
+	int64_t diff = (int64_t)bwg_b->bwg_left_int - (int64_t)bwg_a->bwg_left_int;
+	int64_t diff2 = (int64_t)bwg_b->bwg_left_remainder -
+			(int64_t)bwg_a->bwg_left_remainder;
+	if (diff > 0)
+		return 1;
+	else if (diff < 0)
+		return -1;
+	else if (diff2 > 0)
+		return 1;
+	else if (diff2 < 0)
+		return -1;
+	else
+		return 0;
+}
+
+static void
+bandwidth_left(int port_id, uint64_t load, uint8_t update_idx,
+		struct bwg_slave *bwg_slave)
+{
+	struct rte_eth_link link_status;
+
+	rte_eth_link_get(port_id, &link_status);
+	uint64_t link_bwg = link_status.link_speed * 1000000ULL / 8;
+	if (link_bwg == 0)
+		return;
+	link_bwg = (link_bwg * (update_idx+1) * REORDER_PERIOD_MS);
+	bwg_slave->bwg_left_int = (link_bwg - 1000*load) / link_bwg;
+	bwg_slave->bwg_left_remainder = (link_bwg - 1000*load) % link_bwg;
+}
+
+static void
+bond_ethdev_update_tlb_slave_cb(void *arg)
+{
+	struct bond_dev_private *internals = arg;
+	struct rte_eth_stats slave_stats;
+	struct bwg_slave bwg_array[RTE_MAX_ETHPORTS];
+	uint8_t slave_count;
+	uint64_t tx_bytes;
+
+	uint8_t update_stats = 0;
+	uint8_t i, slave_id;
+
+	internals->slave_update_idx++;
+
+
+	if (internals->slave_update_idx >= REORDER_PERIOD_MS)
+		update_stats = 1;
+
+	for (i = 0; i < internals->active_slave_count; i++) {
+		slave_id = internals->active_slaves[i];
+		rte_eth_stats_get(slave_id, &slave_stats);
+		tx_bytes = slave_stats.obytes - tlb_last_obytets[slave_id];
+		bandwidth_left(slave_id, tx_bytes,
+				internals->slave_update_idx, &bwg_array[i]);
+		bwg_array[i].slave = slave_id;
+
+		if (update_stats)
+			tlb_last_obytets[slave_id] = slave_stats.obytes;
+	}
+
+	if (update_stats == 1)
+		internals->slave_update_idx = 0;
+
+	slave_count = i;
+	qsort(bwg_array, slave_count, sizeof(bwg_array[0]), bandwidth_cmp);
+	for (i = 0; i < slave_count; i++)
+		internals->active_slaves[i] = bwg_array[i].slave;
+
+	rte_eal_alarm_set(REORDER_PERIOD_MS * 1000, bond_ethdev_update_tlb_slave_cb,
+			(struct bond_dev_private *)internals);
+}
+
+static uint16_t
+bond_ethdev_tx_burst_tlb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
+{
+	struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
+	struct bond_dev_private *internals = bd_tx_q->dev_private;
+
+	struct rte_eth_dev *primary_port =
+			&rte_eth_devices[internals->primary_port];
+	uint16_t num_tx_total = 0;
+	uint8_t i, j;
+
+	uint8_t num_of_slaves = internals->active_slave_count;
+	uint8_t slaves[RTE_MAX_ETHPORTS];
+
+	struct ether_hdr *ether_hdr;
+	struct ether_addr primary_slave_addr;
+	struct ether_addr active_slave_addr;
+
+	if (num_of_slaves < 1)
+		return num_tx_total;
+
+	memcpy(slaves, internals->active_slaves,
+				sizeof(internals->active_slaves[0]) * num_of_slaves);
+
+
+	ether_addr_copy(primary_port->data->mac_addrs, &primary_slave_addr);
+
+	if (nb_pkts > 3) {
+		for (i = 0; i < 3; i++)
+			rte_prefetch0(rte_pktmbuf_mtod(bufs[i], void*));
+	}
+
+	for (i = 0; i < num_of_slaves; i++) {
+		ether_addr_copy(&internals->slaves[slaves[i]].persisted_mac_addr,
+				&active_slave_addr);
+
+		for (j = num_tx_total; j < nb_pkts; j++) {
+			if (j + 3 < nb_pkts)
+				rte_prefetch0(rte_pktmbuf_mtod(bufs[j+3], void*));
+
+			ether_hdr = rte_pktmbuf_mtod(bufs[j], struct ether_hdr *);
+			if (is_same_ether_addr(&ether_hdr->s_addr, &primary_slave_addr))
+				ether_addr_copy(&active_slave_addr, &ether_hdr->s_addr);
+		}
+
+		num_tx_total += rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
+				bufs + num_tx_total, nb_pkts - num_tx_total);
+
+		if (num_tx_total == nb_pkts)
+			break;
+	}
+
+	return num_tx_total;
+}
+
 static uint16_t
 bond_ethdev_tx_burst_balance(void *queue, struct rte_mbuf **bufs,
 		uint16_t nb_pkts)
@@ -500,6 +643,7 @@ mac_address_slaves_update(struct rte_eth_dev *bonded_eth_dev)
 		}
 		break;
 	case BONDING_MODE_ACTIVE_BACKUP:
+	case BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING:
 	default:
 		for (i = 0; i < internals->slave_count; i++) {
 			if (internals->slaves[i].port_id ==
@@ -551,6 +695,10 @@ bond_ethdev_mode_set(struct rte_eth_dev *eth_dev, int mode)
 		eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
 		break;
 #endif
+	case BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING:
+		eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_tlb;
+		eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
+		break;
 	default:
 		return -1;
 	}
@@ -676,7 +824,7 @@ slave_add(struct bond_dev_private *internals,
 	}
 
 	slave_details->link_status_wait_to_complete = 0;
-
+	/* clean tlb_last_obytes when adding port for bonding device */
 	memcpy(&(slave_details->persisted_mac_addr), slave_eth_dev->data->mac_addrs,
 			sizeof(struct ether_addr));
 }
@@ -762,6 +910,9 @@ bond_ethdev_start(struct rte_eth_dev *eth_dev)
 	if (internals->user_defined_primary_port)
 		bond_ethdev_primary_set(internals, internals->primary_port);
 
+	if (internals->mode == BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING)
+		bond_ethdev_update_tlb_slave_cb(internals);
+
 	return 0;
 }
 
@@ -770,6 +921,9 @@ bond_ethdev_stop(struct rte_eth_dev *eth_dev)
 {
 	struct bond_dev_private *internals = eth_dev->data->dev_private;
 
+	if (internals->mode == BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING) {
+		rte_eal_alarm_cancel(bond_ethdev_update_tlb_slave_cb, internals);
+    }
 	internals->active_slave_count = 0;
 	internals->link_status_polling_enabled = 0;
 
@@ -1016,6 +1170,7 @@ bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev)
 		break;
 	/* Promiscuous mode is propagated only to primary slave */
 	case BONDING_MODE_ACTIVE_BACKUP:
+	case BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING:
 	default:
 		rte_eth_promiscuous_enable(internals->current_primary_port);
 
@@ -1042,6 +1197,7 @@ bond_ethdev_promiscuous_disable(struct rte_eth_dev *dev)
 		break;
 	/* Promiscuous mode is propagated only to primary slave */
 	case BONDING_MODE_ACTIVE_BACKUP:
+	case BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING:
 	default:
 		rte_eth_promiscuous_disable(internals->current_primary_port);
 	}
diff --git a/lib/librte_pmd_bond/rte_eth_bond_private.h b/lib/librte_pmd_bond/rte_eth_bond_private.h
index 6254c84..2a4e129 100644
--- a/lib/librte_pmd_bond/rte_eth_bond_private.h
+++ b/lib/librte_pmd_bond/rte_eth_bond_private.h
@@ -102,7 +102,6 @@ struct bond_slave_details {
 	uint8_t link_status_poll_enabled;
 	uint8_t link_status_wait_to_complete;
 	uint8_t last_link_status;
-
 	/**< Port Id of slave eth_dev */
 	struct ether_addr persisted_mac_addr;
 };
@@ -145,6 +144,7 @@ struct bond_dev_private {
 	/**< Arary of bonded slaves details */
 
 	struct rte_kvargs *kvlist;
+	uint8_t slave_update_idx;
 };
 
 extern struct eth_dev_ops default_dev_ops;
-- 
2.1.0

  reply	other threads:[~2014-11-27 16:33 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-11-27 16:33 [dpdk-dev] [PATCH v4 0/2] ADD mode 5(tlb) " Daniel Mrzyglod
2014-11-27 16:33 ` Daniel Mrzyglod [this message]
2014-11-27 16:33 ` [dpdk-dev] [PATCH v4 2/2] Unit tests for Mode 5 of Bonding Transmit Load balancig Daniel Mrzyglod
2014-11-27 16:41 ` [dpdk-dev] [PATCH v4 0/2] ADD mode 5(tlb) to link bonding pmd Thomas Monjalon
2014-11-27 16:56   ` Doherty, Declan
2014-11-27 20:27 ` Thomas Monjalon

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1417106021-22842-2-git-send-email-danielx.t.mrzyglod@intel.com \
    --to=danielx.t.mrzyglod@intel.com \
    --cc=dev@dpdk.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).