dpif-netdev: Change rxq_scheduling to use rxq processing cycles.

Previously rxqs were assigned to pmds by round robin in
port/queue order.

Now that we have the processing cycles used for existing rxqs,
use that information to try and produced a better balanced
distribution of rxqs across pmds. i.e. given multiple pmds, the
rxqs which have consumed the largest amount of processing cycles
will be placed on different pmds.

The rxqs are sorted by their processing cycles and assigned (in
sorted order) round robin across pmds.

Signed-off-by: Kevin Traynor <ktraynor@redhat.com>
Signed-off-by: Darrell Ball <dlu998@gmail.com>

diff --git a/Documentation/howto/dpdk.rst b/Documentation/howto/dpdk.rst
index d7f6610bea5c766f3f314fb66f3bc4d08f839fb8..a67f3a1facffc745529e761c96d0bde9d065f5c9 100644 (file)
--- a/Documentation/howto/dpdk.rst
+++ b/Documentation/howto/dpdk.rst
@@ -118,6 +118,13 @@ After that PMD threads on cores where RX queues was pinned will become
   ``core_id`` not in ``pmd-cpu-mask``), RX queue will not be polled by any PMD
   thread.
 
+If pmd-rxq-affinity is not set for rxqs, they will be assigned to pmds (cores)
+automatically. The processing cycles that have been stored for each rxq
+will be used where known to assign rxqs to pmd based on a round robin of the
+sorted rxqs.
+
+Rxq to pmds assignment takes place whenever there are configuration changes.
+
 QoS
 ---
 

diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index 9649e85ff883d1fb7733c2c38ac1074d9934556c..7331b0bdb08d7476724f276f30dd9d688e7a884e 100644 (file)
--- a/lib/dpif-netdev.c
+++ b/lib/dpif-netdev.c
@@ -714,6 +714,8 @@ dp_netdev_rxq_get_cycles(struct dp_netdev_rxq *rx,
 static void
 dp_netdev_rxq_set_intrvl_cycles(struct dp_netdev_rxq *rx,
                            unsigned long long cycles);
+static uint64_t
+dp_netdev_rxq_get_intrvl_cycles(struct dp_netdev_rxq *rx, unsigned idx);
 static void
 dpif_netdev_xps_revalidate_pmd(const struct dp_netdev_pmd_thread *pmd,
                                long long now, bool purge);
@@ -3168,6 +3170,14 @@ dp_netdev_rxq_set_intrvl_cycles(struct dp_netdev_rxq *rx,
                                            % PMD_RXQ_INTERVAL_MAX], cycles);
 }
 
+static uint64_t
+dp_netdev_rxq_get_intrvl_cycles(struct dp_netdev_rxq *rx, unsigned idx)
+{
+    unsigned long long processing_cycles;
+    atomic_read_relaxed(&rx->cycles_intrvl[idx], &processing_cycles);
+    return processing_cycles;
+}
+
 static int
 dp_netdev_process_rxq_port(struct dp_netdev_pmd_thread *pmd,
                            struct netdev_rxq *rx,
@@ -3354,10 +3364,39 @@ rr_numa_list_destroy(struct rr_numa_list *rr)
     hmap_destroy(&rr->numas);
 }
 
+/* Sort Rx Queues by the processing cycles they are consuming. */
+static int
+rxq_cycle_sort(const void *a, const void *b)
+{
+    struct dp_netdev_rxq * qa;
+    struct dp_netdev_rxq * qb;
+    uint64_t total_qa, total_qb;
+    unsigned i;
+
+    qa = *(struct dp_netdev_rxq **) a;
+    qb = *(struct dp_netdev_rxq **) b;
+
+    total_qa = total_qb = 0;
+    for (i = 0; i < PMD_RXQ_INTERVAL_MAX; i++) {
+        total_qa += dp_netdev_rxq_get_intrvl_cycles(qa, i);
+        total_qb += dp_netdev_rxq_get_intrvl_cycles(qb, i);
+    }
+    dp_netdev_rxq_set_cycles(qa, RXQ_CYCLES_PROC_HIST, total_qa);
+    dp_netdev_rxq_set_cycles(qb, RXQ_CYCLES_PROC_HIST, total_qb);
+
+    if (total_qa >= total_qb) {
+        return -1;
+    }
+    return 1;
+}
+
 /* Assign pmds to queues.  If 'pinned' is true, assign pmds to pinned
  * queues and marks the pmds as isolated.  Otherwise, assign non isolated
  * pmds to unpinned queues.
  *
+ * If 'pinned' is false queues will be sorted by processing cycles they are
+ * consuming and then assigned to pmds in round robin order.
+ *
  * The function doesn't touch the pmd threads, it just stores the assignment
  * in the 'pmd' member of each rxq. */
 static void
@@ -3366,20 +3405,16 @@ rxq_scheduling(struct dp_netdev *dp, bool pinned) OVS_REQUIRES(dp->port_mutex)
     struct dp_netdev_port *port;
     struct rr_numa_list rr;
     struct rr_numa *non_local_numa = NULL;
-
-    rr_numa_list_populate(dp, &rr);
+    struct dp_netdev_rxq ** rxqs = NULL;
+    int i, n_rxqs = 0;
+    struct rr_numa *numa = NULL;
+    int numa_id;
 
     HMAP_FOR_EACH (port, node, &dp->ports) {
-        struct rr_numa *numa;
-        int numa_id;
-
         if (!netdev_is_pmd(port->netdev)) {
             continue;
         }
 
-        numa_id = netdev_get_numa_id(port->netdev);
-        numa = rr_numa_list_lookup(&rr, numa_id);
-
         for (int qid = 0; qid < port->n_rxq; qid++) {
             struct dp_netdev_rxq *q = &port->rxqs[qid];
 
@@ -3397,36 +3432,62 @@ rxq_scheduling(struct dp_netdev *dp, bool pinned) OVS_REQUIRES(dp->port_mutex)
                     dp_netdev_pmd_unref(pmd);
                 }
             } else if (!pinned && q->core_id == OVS_CORE_UNSPEC) {
-                if (!numa) {
-                    /* There are no pmds on the queue's local NUMA node.
-                       Round-robin on the NUMA nodes that do have pmds. */
-                    non_local_numa = rr_numa_list_next(&rr, non_local_numa);
-                    if (!non_local_numa) {
-                        VLOG_ERR("There is no available (non-isolated) pmd "
-                                 "thread for port \'%s\' queue %d. This queue "
-                                 "will not be polled. Is pmd-cpu-mask set to "
-                                 "zero? Or are all PMDs isolated to other "
-                                 "queues?", netdev_get_name(port->netdev),
-                                 qid);
-                        continue;
-                    }
-                    q->pmd = rr_numa_get_pmd(non_local_numa);
-                    VLOG_WARN("There's no available (non-isolated) pmd thread "
-                              "on numa node %d. Queue %d on port \'%s\' will "
-                              "be assigned to the pmd on core %d "
-                              "(numa node %d). Expect reduced performance.",
-                              numa_id, qid, netdev_get_name(port->netdev),
-                              q->pmd->core_id, q->pmd->numa_id);
+                if (n_rxqs == 0) {
+                    rxqs = xmalloc(sizeof *rxqs);
                 } else {
-                    /* Assign queue to the next (round-robin) PMD on it's local
-                       NUMA node. */
-                    q->pmd = rr_numa_get_pmd(numa);
+                    rxqs = xrealloc(rxqs, sizeof *rxqs * (n_rxqs + 1));
                 }
+                /* Store the queue. */
+                rxqs[n_rxqs++] = q;
+            }
+        }
+    }
+
+    if (n_rxqs > 1) {
+        /* Sort the queues in order of the processing cycles
+         * they consumed during their last pmd interval. */
+        qsort(rxqs, n_rxqs, sizeof *rxqs, rxq_cycle_sort);
+    }
+
+    rr_numa_list_populate(dp, &rr);
+    /* Assign the sorted queues to pmds in round robin. */
+    for (i = 0; i < n_rxqs; i++) {
+        numa_id = netdev_get_numa_id(rxqs[i]->port->netdev);
+        numa = rr_numa_list_lookup(&rr, numa_id);
+        if (!numa) {
+            /* There are no pmds on the queue's local NUMA node.
+               Round robin on the NUMA nodes that do have pmds. */
+            non_local_numa = rr_numa_list_next(&rr, non_local_numa);
+            if (!non_local_numa) {
+                VLOG_ERR("There is no available (non-isolated) pmd "
+                         "thread for port \'%s\' queue %d. This queue "
+                         "will not be polled. Is pmd-cpu-mask set to "
+                         "zero? Or are all PMDs isolated to other "
+                         "queues?", netdev_rxq_get_name(rxqs[i]->rx),
+                         netdev_rxq_get_queue_id(rxqs[i]->rx));
+                continue;
             }
+            rxqs[i]->pmd = rr_numa_get_pmd(non_local_numa);
+            VLOG_WARN("There's no available (non-isolated) pmd thread "
+                      "on numa node %d. Queue %d on port \'%s\' will "
+                      "be assigned to the pmd on core %d "
+                      "(numa node %d). Expect reduced performance.",
+                      numa_id, netdev_rxq_get_queue_id(rxqs[i]->rx),
+                      netdev_rxq_get_name(rxqs[i]->rx),
+                      rxqs[i]->pmd->core_id, rxqs[i]->pmd->numa_id);
+        } else {
+        rxqs[i]->pmd = rr_numa_get_pmd(numa);
+        VLOG_INFO("Core %d on numa node %d assigned port \'%s\' "
+                  "rx queue %d (measured processing cycles %"PRIu64").",
+                  rxqs[i]->pmd->core_id, numa_id,
+                  netdev_rxq_get_name(rxqs[i]->rx),
+                  netdev_rxq_get_queue_id(rxqs[i]->rx),
+                  dp_netdev_rxq_get_cycles(rxqs[i], RXQ_CYCLES_PROC_HIST));
         }
     }
 
     rr_numa_list_destroy(&rr);
+    free(rxqs);
 }
 
 static void