Initial check-in of kernel-based elephant flow detection.elephant

Areas to work on:

    - Doesn't populate "elephant-flows" field.
    - Doesn't properly handle tunnels.
    - Doesn't have clean way to query elephant table.
    - Double-check locking.
    - Should use names instead of number for mechanism.
    - When changing detection mechanism, should clear old table.
    - Breaks unit tests

3 files changed, 116 insertions, 2 deletions

diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c
index 6dcc2b878..f85f9b740 100644
--- a/vswitchd/bridge.c
+++ b/vswitchd/bridge.c
@@ -228,6 +228,7 @@ static void bridge_configure_mcast_snooping(struct bridge *);
 static void bridge_configure_sflow(struct bridge *, int *sflow_bridge_number);
 static void bridge_configure_ipfix(struct bridge *);
 static void bridge_configure_stp(struct bridge *);
+static void bridge_configure_elephant(struct bridge *);
 static void bridge_configure_tables(struct bridge *);
 static void bridge_configure_dp_desc(struct bridge *);
 static void bridge_configure_remotes(struct bridge *,
@@ -266,6 +267,8 @@ static void mirror_destroy(struct mirror *);
 static bool mirror_configure(struct mirror *);
 static void mirror_refresh_stats(struct mirror *);
 
+static void elephants_refresh_stats(struct bridge *);
+
 static void iface_configure_lacp(struct iface *, struct lacp_slave_settings *);
 static bool iface_create(struct bridge *, const struct ovsrec_interface *,
                          const struct ovsrec_port *);
@@ -378,6 +381,7 @@ bridge_init(const char *remote)
 
     ovsdb_idl_omit_alert(idl, &ovsrec_bridge_col_datapath_id);
     ovsdb_idl_omit_alert(idl, &ovsrec_bridge_col_status);
+    ovsdb_idl_omit_alert(idl, &ovsrec_bridge_col_elephant_flows);
     ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
 
     ovsdb_idl_omit_alert(idl, &ovsrec_port_col_status);
@@ -622,6 +626,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
         bridge_configure_sflow(br, &sflow_bridge_number);
         bridge_configure_ipfix(br);
         bridge_configure_stp(br);
+        bridge_configure_elephant(br);
         bridge_configure_tables(br);
         bridge_configure_dp_desc(br);
     }
@@ -1383,6 +1388,29 @@ bridge_configure_stp(struct bridge *br)
     }
 }
 
+/* Set elephant flow detection configuration on 'br'. */
+static void
+bridge_configure_elephant(struct bridge *br)
+{
+    const char *config_str;
+    uint64_t mech, arg1, arg2;
+    int dscp;
+
+    config_str = smap_get(&br->cfg->other_config, "elephant-mech");
+    mech = config_str ? strtoul(config_str, NULL, 0) : 0;
+
+    config_str = smap_get(&br->cfg->other_config, "elephant-arg1");
+    arg1 = config_str ? strtoul(config_str, NULL, 0) : 0;
+
+    config_str = smap_get(&br->cfg->other_config, "elephant-arg2");
+    arg2 = config_str ? strtoul(config_str, NULL, 0) : 0;
+
+    config_str = smap_get(&br->cfg->other_config, "elephant-dscp");
+    dscp = config_str ? strtoul(config_str, NULL, 0) : -1;
+
+    ofproto_set_elephant(br->ofproto, mech, arg1, arg2, dscp);
+}
+
 static bool
 bridge_has_bond_fake_iface(const struct bridge *br, const char *name)
 {
@@ -2459,6 +2487,8 @@ bridge_run(void)
                     mirror_refresh_stats(m);
                 }
 
+                /* xxx Find better place for this, but useful timer. */
+                elephants_refresh_stats(br);
             }
             refresh_controller_status();
             ovsdb_idl_txn_commit(txn);
@@ -4337,3 +4367,15 @@ mirror_refresh_stats(struct mirror *m)
 
     ovsrec_mirror_set_statistics(m->cfg, keys, values, stat_cnt);
 }
+
+static void
+elephants_refresh_stats(struct bridge *br)
+{
+    struct smap elephants;
+
+    if (!ofproto_get_elephants(br->ofproto, &elephants)) {
+        ovsrec_bridge_set_elephant_flows(br->cfg, &elephants);
+    }
+
+    smap_destroy(&elephants);
+}
diff --git a/vswitchd/vswitch.ovsschema b/vswitchd/vswitch.ovsschema
index bc9ea73c6..717e61506 100644
--- a/vswitchd/vswitch.ovsschema
+++ b/vswitchd/vswitch.ovsschema
@@ -1,6 +1,6 @@
 {"name": "Open_vSwitch",
- "version": "7.8.0",
- "cksum": "2676751133 20740",
+ "version": "7.78.0",
+ "cksum": "904547257 20882",
  "tables": {
    "Open_vSwitch": {
      "columns": {
@@ -93,6 +93,9 @@
          "type": {"key": {"type": "string",
                           "enum": ["set", ["standalone", "secure"]]},
                   "min": 0, "max": 1}},
+       "elephant_flows": {
+         "type": {"key": "string", "value": "string", "min": 0, "max": "unlimited"},
+         "ephemeral": true},
        "status": {
          "type": {"key": "string", "value": "string", "min": 0, "max": "unlimited"},
          "ephemeral": true},
diff --git a/vswitchd/vswitch.xml b/vswitchd/vswitch.xml
index d47fc1a6e..18cf7db35 100644
--- a/vswitchd/vswitch.xml
+++ b/vswitchd/vswitch.xml
@@ -480,6 +480,75 @@
       </column>
     </group>
 
+    <group title="Elephant Flow Detection">
+      <p>
+        The conventional wisdom is that majority of flows in the
+        datacenter are short (mice), yet the majority of packets belong to
+        a few long-lived flows (elephants).  Mice are often associated
+        with bursty, latency sensitive apps whereas elephants tend to be
+        large transfers in which throughput is far more important than
+        latency.
+      </p>
+
+      <p>
+        Open vSwitch has the ability to distinguish elephant flows and
+        identify them through the database or mark them.  Elephants are
+        identified by the combination of their duration and number of
+        bytes.
+      </p>
+
+      <column name="other_config" key="elephant-mech">
+        The mechanism used to detect elephants.  It may be set to one of
+        the following:
+        <dl>
+          <dt><code>0</code></dt>
+          <dd>Disable elephant detection.</dd>
+
+          <dt><code>1</code></dt>
+          <dd>Bytes.  A flow is declared an elephant based on a
+          threshold of the total number of bytes sent.  <ref
+          column="other_config" key="elephant-arg1"/> specifies the
+          number of bytes.  <ref column="other_config"
+          key="elephant-arg2"/> specifies the number of seconds that the
+          flow must exist before being declared an elephant; set to
+          <code>0</code> for immediate detection when the byte threshold
+          is crossed.</dd>
+
+          <dt><code>2</code></dt>
+          <dd>TSO.  A flow is declared an elephant based on the segment
+          size.  <ref column="other_config" key="elephant-arg1"/>
+          specifies the size of the segment.  <ref column="other_config"
+          key="elephant-arg2"/> specifies the required number of
+          segments of the configured size that must be seen.
+          </dd>
+        </dl>
+      </column>
+
+      <column name="other_config" key="elephant-arg1">
+        First argument to elephant mechanism.  The arguments are
+        described in <ref column="other_config" key="elephant-mech"/>.
+      </column>
+
+      <column name="other_config" key="elephant-arg2">
+        Second argument to elephant mechanism.  The arguments are
+        described in <ref column="other_config" key="elephant-mech"/>.
+      </column>
+
+      <column name="other_config" key="elephant-dscp">
+        When an elephant flow is detected sets the DSCP value to use on
+        the wire for packets matching that flow.  Not setting the key or
+        using a value of -1 will leave elephant flows unmarked.
+      </column>
+
+      <column name="elephant_flows">
+        <code>xxx Not currently supported xxx</code>
+        The <code>elephants</code> column contains key-value pairs that
+        report elephant flows.  The key is a description of the flow,
+        and the value is size of the flow in bytes.  These are updated
+        periodically (currently, every 5 seconds).
+      </column>
+    </group>
+
     <group title="OpenFlow Configuration">
       <column name="controller">
         <p>