bond: performance enhancement
We were only puting one packet per frame to the output node. Change to
buffer multiple packets per frame. Performance is now on top of dpdk-based
bonding.
Put a spinlock in the tx thread in case the rug is pulled under us.
Change-Id: Ifda5af086a984a7301972cd6c8e428217f676a95
Signed-off-by: Steven <sluong@cisco.com>
diff --git a/src/vnet/bonding/device.c b/src/vnet/bonding/device.c
index 8f9b3a9..cebfd69 100644
--- a/src/vnet/bonding/device.c
+++ b/src/vnet/bonding/device.c
@@ -351,9 +351,9 @@
uword n_trace = vlib_get_trace_count (vm, node);
u16 thread_index = vlib_get_thread_index ();
vnet_main_t *vnm = vnet_get_main ();
- u32 *to_next, *to_next1, *to_next2, *to_next3;
+ u32 *to_next;
u32 sif_if_index, sif_if_index1, sif_if_index2, sif_if_index3;
- vlib_frame_t *f, *f1, *f2, *f3;
+ vlib_frame_t *f;
if (PREDICT_FALSE (bif->admin_up == 0))
{
@@ -367,6 +367,7 @@
return frame->n_vectors;
}
+ clib_spinlock_lock_if_init (&bif->lockp);
if (PREDICT_FALSE (vec_len (bif->active_slaves) == 0))
{
bi0 = from[0];
@@ -383,206 +384,245 @@
frame->n_vectors);
vlib_error_count (vm, node->node_index, BOND_TX_ERROR_NO_SLAVE,
frame->n_vectors);
+ clib_spinlock_unlock_if_init (&bif->lockp);
return frame->n_vectors;
}
+ vec_validate_aligned (bif->per_thread_info[thread_index].frame,
+ vec_len (bif->active_slaves), CLIB_CACHE_LINE_BYTES);
+
/* Number of buffers / pkts */
n_left_from = frame->n_vectors;
- while (n_left_from >= 8)
+ while (n_left_from > 0)
{
- // Prefetch next iteration
- {
- vlib_buffer_t *p4, *p5, *p6, *p7;
-
- p4 = vlib_get_buffer (vm, from[4]);
- p5 = vlib_get_buffer (vm, from[5]);
- p6 = vlib_get_buffer (vm, from[6]);
- p7 = vlib_get_buffer (vm, from[7]);
-
- vlib_prefetch_buffer_header (p4, STORE);
- vlib_prefetch_buffer_header (p5, STORE);
- vlib_prefetch_buffer_header (p6, STORE);
- vlib_prefetch_buffer_header (p7, STORE);
-
- CLIB_PREFETCH (p4->data, CLIB_CACHE_LINE_BYTES, LOAD);
- CLIB_PREFETCH (p5->data, CLIB_CACHE_LINE_BYTES, LOAD);
- CLIB_PREFETCH (p6->data, CLIB_CACHE_LINE_BYTES, LOAD);
- CLIB_PREFETCH (p7->data, CLIB_CACHE_LINE_BYTES, LOAD);
- }
-
- bi0 = from[0];
- bi1 = from[1];
- bi2 = from[2];
- bi3 = from[3];
-
- b0 = vlib_get_buffer (vm, bi0);
- b1 = vlib_get_buffer (vm, bi1);
- b2 = vlib_get_buffer (vm, bi2);
- b3 = vlib_get_buffer (vm, bi3);
-
- VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b0);
- VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b1);
- VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b2);
- VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b3);
-
- sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_TX];
- sw_if_index1 = vnet_buffer (b1)->sw_if_index[VLIB_TX];
- sw_if_index2 = vnet_buffer (b2)->sw_if_index[VLIB_TX];
- sw_if_index3 = vnet_buffer (b3)->sw_if_index[VLIB_TX];
-
- port =
- (bond_load_balance_table[bif->lb]).load_balance (vm, node, bif, b0);
- port1 =
- (bond_load_balance_table[bif->lb]).load_balance (vm, node, bif, b1);
- port2 =
- (bond_load_balance_table[bif->lb]).load_balance (vm, node, bif, b2);
- port3 =
- (bond_load_balance_table[bif->lb]).load_balance (vm, node, bif, b3);
-
- sif_if_index = *vec_elt_at_index (bif->active_slaves, port);
- sif_if_index1 = *vec_elt_at_index (bif->active_slaves, port1);
- sif_if_index2 = *vec_elt_at_index (bif->active_slaves, port2);
- sif_if_index3 = *vec_elt_at_index (bif->active_slaves, port3);
-
- vnet_buffer (b0)->sw_if_index[VLIB_TX] = sif_if_index;
- vnet_buffer (b1)->sw_if_index[VLIB_TX] = sif_if_index1;
- vnet_buffer (b2)->sw_if_index[VLIB_TX] = sif_if_index2;
- vnet_buffer (b3)->sw_if_index[VLIB_TX] = sif_if_index3;
-
- f = vnet_get_frame_to_sw_interface (vnm, sif_if_index);
- f1 = vnet_get_frame_to_sw_interface (vnm, sif_if_index1);
- f2 = vnet_get_frame_to_sw_interface (vnm, sif_if_index2);
- f3 = vnet_get_frame_to_sw_interface (vnm, sif_if_index3);
-
- to_next = vlib_frame_vector_args (f);
- to_next1 = vlib_frame_vector_args (f1);
- to_next2 = vlib_frame_vector_args (f2);
- to_next3 = vlib_frame_vector_args (f3);
-
- to_next += f->n_vectors;
- to_next1 += f1->n_vectors;
- to_next2 += f2->n_vectors;
- to_next3 += f3->n_vectors;
-
- to_next[0] = vlib_get_buffer_index (vm, b0);
- to_next1[0] = vlib_get_buffer_index (vm, b1);
- to_next2[0] = vlib_get_buffer_index (vm, b2);
- to_next3[0] = vlib_get_buffer_index (vm, b3);
-
- f->n_vectors++;
- f1->n_vectors++;
- f2->n_vectors++;
- f3->n_vectors++;
-
- vnet_put_frame_to_sw_interface (vnm, sif_if_index, f);
- vnet_put_frame_to_sw_interface (vnm, sif_if_index1, f1);
- vnet_put_frame_to_sw_interface (vnm, sif_if_index2, f2);
- vnet_put_frame_to_sw_interface (vnm, sif_if_index3, f3);
-
- if (PREDICT_FALSE (n_trace > 0))
+ while (n_left_from >= 4)
{
- vlib_trace_buffer (vm, node, next0, b0, 0 /* follow_chain */ );
- vlib_set_trace_count (vm, node, --n_trace);
- t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
- eth = (ethernet_header_t *) vlib_buffer_get_current (b0);
- t0->ethernet = *eth;
- t0->sw_if_index = sw_if_index;
- t0->bond_sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_TX];
-
- if (PREDICT_TRUE (n_trace > 0))
+ // Prefetch next iteration
+ if (n_left_from >= 8)
{
- vlib_trace_buffer (vm, node, next1, b1, 0 /* follow_chain */ );
+ vlib_buffer_t *p4, *p5, *p6, *p7;
+
+ p4 = vlib_get_buffer (vm, from[4]);
+ p5 = vlib_get_buffer (vm, from[5]);
+ p6 = vlib_get_buffer (vm, from[6]);
+ p7 = vlib_get_buffer (vm, from[7]);
+
+ vlib_prefetch_buffer_header (p4, STORE);
+ vlib_prefetch_buffer_header (p5, STORE);
+ vlib_prefetch_buffer_header (p6, STORE);
+ vlib_prefetch_buffer_header (p7, STORE);
+
+ CLIB_PREFETCH (p4->data, CLIB_CACHE_LINE_BYTES, LOAD);
+ CLIB_PREFETCH (p5->data, CLIB_CACHE_LINE_BYTES, LOAD);
+ CLIB_PREFETCH (p6->data, CLIB_CACHE_LINE_BYTES, LOAD);
+ CLIB_PREFETCH (p7->data, CLIB_CACHE_LINE_BYTES, LOAD);
+ }
+
+ bi0 = from[0];
+ bi1 = from[1];
+ bi2 = from[2];
+ bi3 = from[3];
+
+ b0 = vlib_get_buffer (vm, bi0);
+ b1 = vlib_get_buffer (vm, bi1);
+ b2 = vlib_get_buffer (vm, bi2);
+ b3 = vlib_get_buffer (vm, bi3);
+
+ VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b0);
+ VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b1);
+ VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b2);
+ VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b3);
+
+ sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_TX];
+ sw_if_index1 = vnet_buffer (b1)->sw_if_index[VLIB_TX];
+ sw_if_index2 = vnet_buffer (b2)->sw_if_index[VLIB_TX];
+ sw_if_index3 = vnet_buffer (b3)->sw_if_index[VLIB_TX];
+
+ port =
+ (bond_load_balance_table[bif->lb]).load_balance (vm, node, bif,
+ b0);
+ port1 =
+ (bond_load_balance_table[bif->lb]).load_balance (vm, node, bif,
+ b1);
+ port2 =
+ (bond_load_balance_table[bif->lb]).load_balance (vm, node, bif,
+ b2);
+ port3 =
+ (bond_load_balance_table[bif->lb]).load_balance (vm, node, bif,
+ b3);
+
+ sif_if_index = *vec_elt_at_index (bif->active_slaves, port);
+ sif_if_index1 = *vec_elt_at_index (bif->active_slaves, port1);
+ sif_if_index2 = *vec_elt_at_index (bif->active_slaves, port2);
+ sif_if_index3 = *vec_elt_at_index (bif->active_slaves, port3);
+
+ vnet_buffer (b0)->sw_if_index[VLIB_TX] = sif_if_index;
+ vnet_buffer (b1)->sw_if_index[VLIB_TX] = sif_if_index1;
+ vnet_buffer (b2)->sw_if_index[VLIB_TX] = sif_if_index2;
+ vnet_buffer (b3)->sw_if_index[VLIB_TX] = sif_if_index3;
+
+ if (bif->per_thread_info[thread_index].frame[port] == 0)
+ bif->per_thread_info[thread_index].frame[port] =
+ vnet_get_frame_to_sw_interface (vnm, sif_if_index);
+
+ if (bif->per_thread_info[thread_index].frame[port1] == 0)
+ bif->per_thread_info[thread_index].frame[port1] =
+ vnet_get_frame_to_sw_interface (vnm, sif_if_index1);
+
+ if (bif->per_thread_info[thread_index].frame[port2] == 0)
+ bif->per_thread_info[thread_index].frame[port2] =
+ vnet_get_frame_to_sw_interface (vnm, sif_if_index2);
+
+ if (bif->per_thread_info[thread_index].frame[port3] == 0)
+ bif->per_thread_info[thread_index].frame[port3] =
+ vnet_get_frame_to_sw_interface (vnm, sif_if_index3);
+
+ f = bif->per_thread_info[thread_index].frame[port];
+ to_next = vlib_frame_vector_args (f);
+ to_next += f->n_vectors;
+ to_next[0] = vlib_get_buffer_index (vm, b0);
+ f->n_vectors++;
+
+ f = bif->per_thread_info[thread_index].frame[port1];
+ to_next = vlib_frame_vector_args (f);
+ to_next += f->n_vectors;
+ to_next[0] = vlib_get_buffer_index (vm, b1);
+ f->n_vectors++;
+
+ f = bif->per_thread_info[thread_index].frame[port2];
+ to_next = vlib_frame_vector_args (f);
+ to_next += f->n_vectors;
+ to_next[0] = vlib_get_buffer_index (vm, b2);
+ f->n_vectors++;
+
+ f = bif->per_thread_info[thread_index].frame[port3];
+ to_next = vlib_frame_vector_args (f);
+ to_next += f->n_vectors;
+ to_next[0] = vlib_get_buffer_index (vm, b3);
+ f->n_vectors++;
+
+ if (PREDICT_FALSE (n_trace > 0))
+ {
+ vlib_trace_buffer (vm, node, next0, b0, 0 /* follow_chain */ );
vlib_set_trace_count (vm, node, --n_trace);
- t0 = vlib_add_trace (vm, node, b1, sizeof (*t0));
- eth = (ethernet_header_t *) vlib_buffer_get_current (b1);
+ t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
+ eth = (ethernet_header_t *) vlib_buffer_get_current (b0);
t0->ethernet = *eth;
- t0->sw_if_index = sw_if_index1;
- t0->bond_sw_if_index = vnet_buffer (b1)->sw_if_index[VLIB_TX];
+ t0->sw_if_index = sw_if_index;
+ t0->bond_sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_TX];
if (PREDICT_TRUE (n_trace > 0))
{
- vlib_trace_buffer (vm, node, next2, b2,
+ vlib_trace_buffer (vm, node, next1, b1,
0 /* follow_chain */ );
vlib_set_trace_count (vm, node, --n_trace);
- t0 = vlib_add_trace (vm, node, b2, sizeof (*t0));
- eth = (ethernet_header_t *) vlib_buffer_get_current (b2);
+ t0 = vlib_add_trace (vm, node, b1, sizeof (*t0));
+ eth = (ethernet_header_t *) vlib_buffer_get_current (b1);
t0->ethernet = *eth;
- t0->sw_if_index = sw_if_index2;
+ t0->sw_if_index = sw_if_index1;
t0->bond_sw_if_index =
- vnet_buffer (b2)->sw_if_index[VLIB_TX];
+ vnet_buffer (b1)->sw_if_index[VLIB_TX];
if (PREDICT_TRUE (n_trace > 0))
{
- vlib_trace_buffer (vm, node, next3, b3,
+ vlib_trace_buffer (vm, node, next2, b2,
0 /* follow_chain */ );
vlib_set_trace_count (vm, node, --n_trace);
- t0 = vlib_add_trace (vm, node, b3, sizeof (*t0));
+ t0 = vlib_add_trace (vm, node, b2, sizeof (*t0));
eth =
- (ethernet_header_t *) vlib_buffer_get_current (b3);
+ (ethernet_header_t *) vlib_buffer_get_current (b2);
t0->ethernet = *eth;
- t0->sw_if_index = sw_if_index3;
+ t0->sw_if_index = sw_if_index2;
t0->bond_sw_if_index =
- vnet_buffer (b3)->sw_if_index[VLIB_TX];
+ vnet_buffer (b2)->sw_if_index[VLIB_TX];
+
+ if (PREDICT_TRUE (n_trace > 0))
+ {
+ vlib_trace_buffer (vm, node, next3, b3,
+ 0 /* follow_chain */ );
+ vlib_set_trace_count (vm, node, --n_trace);
+ t0 = vlib_add_trace (vm, node, b3, sizeof (*t0));
+ eth =
+ (ethernet_header_t *)
+ vlib_buffer_get_current (b3);
+ t0->ethernet = *eth;
+ t0->sw_if_index = sw_if_index3;
+ t0->bond_sw_if_index =
+ vnet_buffer (b3)->sw_if_index[VLIB_TX];
+ }
}
}
}
+ from += 4;
+ n_left_from -= 4;
}
- from += 4;
- n_left_from -= 4;
+ while (n_left_from > 0)
+ {
+ // Prefetch next iteration
+ if (n_left_from > 1)
+ {
+ vlib_buffer_t *p2;
+
+ p2 = vlib_get_buffer (vm, from[1]);
+ vlib_prefetch_buffer_header (p2, STORE);
+ CLIB_PREFETCH (p2->data, CLIB_CACHE_LINE_BYTES, LOAD);
+ }
+
+ bi0 = from[0];
+ b0 = vlib_get_buffer (vm, bi0);
+
+ VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b0);
+
+ sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_TX];
+
+ port =
+ (bond_load_balance_table[bif->lb]).load_balance (vm, node, bif,
+ b0);
+ sif_if_index = *vec_elt_at_index (bif->active_slaves, port);
+ vnet_buffer (b0)->sw_if_index[VLIB_TX] = sif_if_index;
+ if (bif->per_thread_info[thread_index].frame[port] == 0)
+ bif->per_thread_info[thread_index].frame[port] =
+ vnet_get_frame_to_sw_interface (vnm, sif_if_index);
+ f = bif->per_thread_info[thread_index].frame[port];
+ to_next = vlib_frame_vector_args (f);
+ to_next += f->n_vectors;
+ to_next[0] = vlib_get_buffer_index (vm, b0);
+ f->n_vectors++;
+
+ if (PREDICT_FALSE (n_trace > 0))
+ {
+ vlib_trace_buffer (vm, node, next0, b0, 0 /* follow_chain */ );
+ vlib_set_trace_count (vm, node, --n_trace);
+ t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
+ eth = (ethernet_header_t *) vlib_buffer_get_current (b0);
+ t0->ethernet = *eth;
+ t0->sw_if_index = sw_if_index;
+ t0->bond_sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_TX];
+ }
+
+ from += 1;
+ n_left_from -= 1;
+ }
}
- while (n_left_from > 0)
+ for (port = 0; port < vec_len (bif->active_slaves); port++)
{
- // Prefetch next iteration
- if (n_left_from > 1)
- {
- vlib_buffer_t *p2;
+ f = bif->per_thread_info[thread_index].frame[port];
+ if (f == 0)
+ continue;
- p2 = vlib_get_buffer (vm, from[1]);
- vlib_prefetch_buffer_header (p2, STORE);
- CLIB_PREFETCH (p2->data, CLIB_CACHE_LINE_BYTES, LOAD);
- }
-
- bi0 = from[0];
- b0 = vlib_get_buffer (vm, bi0);
-
- VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b0);
-
- sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_TX];
-
- port =
- (bond_load_balance_table[bif->lb]).load_balance (vm, node, bif, b0);
- sif_if_index = *vec_elt_at_index (bif->active_slaves, port);
- vnet_buffer (b0)->sw_if_index[VLIB_TX] = sif_if_index;
- f = vnet_get_frame_to_sw_interface (vnm, sif_if_index);
- to_next = vlib_frame_vector_args (f);
- to_next += f->n_vectors;
-
- to_next[0] = vlib_get_buffer_index (vm, b0);
- f->n_vectors++;
- vnet_put_frame_to_sw_interface (vnm, sif_if_index, f);
-
- if (PREDICT_FALSE (n_trace > 0))
- {
- vlib_trace_buffer (vm, node, next0, b0, 0 /* follow_chain */ );
- vlib_set_trace_count (vm, node, --n_trace);
- t0 = vlib_add_trace (vm, node, b0, sizeof (*t0));
- eth = (ethernet_header_t *) vlib_buffer_get_current (b0);
- t0->ethernet = *eth;
- t0->sw_if_index = sw_if_index;
- t0->bond_sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_TX];
- }
-
- from += 1;
- n_left_from -= 1;
+ sw_if_index = *vec_elt_at_index (bif->active_slaves, port);
+ vnet_put_frame_to_sw_interface (vnm, sw_if_index, f);
+ bif->per_thread_info[thread_index].frame[port] = 0;
}
vlib_increment_simple_counter (vnet_main.interface_main.sw_if_counters
+ VNET_INTERFACE_COUNTER_TX, thread_index,
bif->sw_if_index, frame->n_vectors);
+ clib_spinlock_unlock_if_init (&bif->lockp);
return frame->n_vectors;
}