ofproto-dpif: Run fast internally.

ofproto-dpif is responsible for quite a few book keeping tasks in
addition to handling flow misses.  Many of these tasks (flow
expiration, flow revalidation, etc) can take many hundreds of
milliseconds, during which no misses can be handled.  The ideal
long term solution to this problem, is to isolate flow miss
handling into it's own thread.  However, for now this patch
provides a 5% increase in TCP_CRR performance, and smooths out
results during revalidations.

Signed-off-by: Ethan Jackson <ethan@nicira.com>

diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c
index 0b16bfa..30ef5f9 100644 (file)
--- a/ofproto/ofproto-dpif.c
+++ b/ofproto/ofproto-dpif.c
@@ -598,6 +598,7 @@ static void port_run_fast(struct ofport_dpif *);
 static void port_wait(struct ofport_dpif *);
 static int set_cfm(struct ofport *, const struct cfm_settings *);
 static void ofport_clear_priorities(struct ofport_dpif *);
+static void run_fast_rl(void);
 
 struct dpif_completion {
     struct list list_node;
@@ -1008,6 +1009,7 @@ type_run(const char *type)
                 if (need_revalidate
                     || tag_set_intersects(&revalidate_set, facet->tags)) {
                     facet_revalidate(facet);
+                    run_fast_rl();
                 }
             }
         }
@@ -1075,18 +1077,10 @@ type_run(const char *type)
 }
 
 static int
-type_run_fast(const char *type)
+dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
 {
-    struct dpif_backer *backer;
     unsigned int work;
 
-    backer = shash_find_data(&all_dpif_backers, type);
-    if (!backer) {
-        /* This is not necessarily a problem, since backers are only
-         * created on demand. */
-        return 0;
-    }
-
     /* Handle one or more batches of upcalls, until there's nothing left to do
      * or until we do a fixed total amount of work.
      *
@@ -1097,8 +1091,8 @@ type_run_fast(const char *type)
      * optimizations can make major improvements on some benchmarks and
      * presumably for real traffic as well. */
     work = 0;
-    while (work < FLOW_MISS_MAX_BATCH) {
-        int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
+    while (work < max_batch) {
+        int retval = handle_upcalls(backer, max_batch - work);
         if (retval <= 0) {
             return -retval;
         }
@@ -1108,6 +1102,58 @@ type_run_fast(const char *type)
     return 0;
 }
 
+static int
+type_run_fast(const char *type)
+{
+    struct dpif_backer *backer;
+
+    backer = shash_find_data(&all_dpif_backers, type);
+    if (!backer) {
+        /* This is not necessarily a problem, since backers are only
+         * created on demand. */
+        return 0;
+    }
+
+    return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
+}
+
+static void
+run_fast_rl(void)
+{
+    static long long int port_rl = LLONG_MIN;
+    static unsigned int backer_rl = 0;
+
+    if (time_msec() >= port_rl) {
+        struct ofproto_dpif *ofproto;
+        struct ofport_dpif *ofport;
+
+        HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
+
+            HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
+                port_run_fast(ofport);
+            }
+        }
+        port_rl = time_msec() + 200;
+    }
+
+    /* XXX: We have to be careful not to do too much work in this function.  If
+     * we call dpif_backer_run_fast() too often, or with too large a batch,
+     * performance improves signifcantly, but at a cost.  It's possible for the
+     * number of flows in the datapath to increase without bound, and for poll
+     * loops to take 10s of seconds.   The correct solution to this problem,
+     * long term, is to separate flow miss handling into it's own thread so it
+     * isn't affected by revalidations, and expirations.  Until then, this is
+     * the best we can do. */
+    if (++backer_rl >= 10) {
+        struct shash_node *node;
+
+        backer_rl = 0;
+        SHASH_FOR_EACH (node, &all_dpif_backers) {
+            dpif_backer_run_fast(node->data, 1);
+        }
+    }
+}
+
 static void
 type_wait(const char *type)
 {
@@ -4254,6 +4300,7 @@ update_stats(struct dpif_backer *backer)
             delete_unexpected_flow(ofproto, key, key_len);
             break;
         }
+        run_fast_rl();
     }
     dpif_flow_dump_done(&dump);
 }
@@ -5057,6 +5104,7 @@ push_all_stats(void)
 
         HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
             facet_push_stats(facet);
+            run_fast_rl();
         }
     }
 
@@ -5229,6 +5277,7 @@ subfacet_destroy_batch(struct ofproto_dpif *ofproto,
         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
         subfacets[i]->path = SF_NOT_INSTALLED;
         subfacet_destroy(subfacets[i]);
+        run_fast_rl();
     }
 }