Merged PR #650 (v2015.11 release) onto master

src/Makefile

@@ -209,5 +209,8 @@ mrproper: clean
 	$(E) "RM        $(RMOBJS)"
 	$(Q)-rm -rf $(RMOBJS)
 
-.PHONY: clean $(TESTMODS) $(TESTSCRIPTS)
+benchmarks:
+	$(Q) (scripts/bench.sh)
+
+.PHONY: clean $(TESTMODS) $(TESTSCRIPTS) benchmarks
 

src/apps/bridge/README.md

@@ -45,38 +45,45 @@ The flooding `bridge` app ignores the *config* key of its configuration.
 
 # Learning bridge (apps.bridge.learning)
 
-The learning `bridge` app implements a *learning bridge* using a [Bloom
-filter](https://en.wikipedia.org/wiki/Bloom_filter) to store the set of
-MAC source addresses of packets arriving on each input port. When a
-packet is received it is forwarded to all output ports whose
-corresponding input ports match the packet's destination MAC address.
-When no input port matches, the packet is flooded to all output ports.
-Multicast MAC addresses are always flooded to all output ports associated
-with the input port. The scoping rules according to the split-horizon
-topology apply unchanged.
+The learning `bridge` app implements a *learning bridge* using a
+custom hash table to store the set of MAC source addresses of packets
+arriving on each input port. When a packet is received it is forwarded
+to all output ports whose corresponding input ports match the packet's
+destination MAC address.  When no input port matches, the packet is
+flooded to all output ports.  Multicast MAC addresses are always
+flooded to all output ports associated with the input port. The
+scoping rules according to the split-horizon topology apply unchanged.
 
 ## Configuration
 
 The learning `bridge` app accepts a table as the value of the *config*
 key of its configuration. The following keys are defined:
 
-— Key **mac_table_size**
+— Key **mac_table**
 
-**Optional**. Expected maximum number of MAC addresses to store in each
-per-port Bloom filter. Default is 1000.
+*Optional*. This is a table that defines the characteristics of the
+  MAC table.  The following keys are defined
 
-— Key **fp_rate**
+— Key **size**
 
-**Optional**. Maximum rate of false-positives for look-ups in the Bloom
-filters, provided the number of distinct objects stored in the filter
-does not exceed *mac_table_size*. Default is 0.001.
+*Optional*. The number of MAC addresses to be stored in the
+ table. Default is 256. The size of the table is increased
+ automatically if this limit is reached or if an overflow in one of
+ the hash buckets occurs.  This value is capped by **resize_max**.
 
 — Key **timeout**
 
-**Optional**. Timeout for learned MAC addresses in seconds. Default is
+*Optional*. Timeout for learned MAC addresses in seconds. Default is
 60.
 
 — Key **verbose**
 
-**Optional**. A boolean value. If true, enables the printing of debugging
-output. Default is `false`.
+*Optional*. A boolean value. If true, statistics about table usage is logged during each timeout interval.  Default is `false`.
+
+— Key **copy_on_resize**
+
+*Optional*. A boolean value. If true, the contents of the table is copied to the newly allocated table after a resize operation.  Default is `true`.
+
+— Key **resize_max**
+
+*Optional*. An upper bound for the size of the table. Default is 65536.

src/apps/bridge/README.md.src

@@ -54,38 +54,45 @@ The flooding `bridge` app ignores the *config* key of its configuration.
 
 # Learning bridge (apps.bridge.learning)
 
-The learning `bridge` app implements a *learning bridge* using a [Bloom
-filter](https://en.wikipedia.org/wiki/Bloom_filter) to store the set of
-MAC source addresses of packets arriving on each input port. When a
-packet is received it is forwarded to all output ports whose
-corresponding input ports match the packet's destination MAC address.
-When no input port matches, the packet is flooded to all output ports.
-Multicast MAC addresses are always flooded to all output ports associated
-with the input port. The scoping rules according to the split-horizon
-topology apply unchanged.
+The learning `bridge` app implements a *learning bridge* using a
+custom hash table to store the set of MAC source addresses of packets
+arriving on each input port. When a packet is received it is forwarded
+to all output ports whose corresponding input ports match the packet's
+destination MAC address.  When no input port matches, the packet is
+flooded to all output ports.  Multicast MAC addresses are always
+flooded to all output ports associated with the input port. The
+scoping rules according to the split-horizon topology apply unchanged.
 
 ## Configuration
 
 The learning `bridge` app accepts a table as the value of the *config*
 key of its configuration. The following keys are defined:
 
-— Key **mac_table_size**
+— Key **mac_table**
 
-**Optional**. Expected maximum number of MAC addresses to store in each
-per-port Bloom filter. Default is 1000.
+*Optional*. This is a table that defines the characteristics of the
+  MAC table.  The following keys are defined
 
-— Key **fp_rate**
+— Key **size**
 
-**Optional**. Maximum rate of false-positives for look-ups in the Bloom
-filters, provided the number of distinct objects stored in the filter
-does not exceed *mac_table_size*. Default is 0.001.
+*Optional*. The number of MAC addresses to be stored in the
+ table. Default is 256. The size of the table is increased
+ automatically if this limit is reached or if an overflow in one of
+ the hash buckets occurs.  This value is capped by **resize_max**.
 
 — Key **timeout**
 
-**Optional**. Timeout for learned MAC addresses in seconds. Default is
+*Optional*. Timeout for learned MAC addresses in seconds. Default is
 60.
 
 — Key **verbose**
 
-**Optional**. A boolean value. If true, enables the printing of debugging
-output. Default is `false`.
+*Optional*. A boolean value. If true, statistics about table usage is logged during each timeout interval.  Default is `false`.
+
+— Key **copy_on_resize**
+
+*Optional*. A boolean value. If true, the contents of the table is copied to the newly allocated table after a resize operation.  Default is `true`.
+
+— Key **resize_max**
+
+*Optional*. An upper bound for the size of the table. Default is 65536.

src/apps/bridge/base.lua

@@ -16,23 +16,46 @@
 --              ...},
 --            config = { <bridge-specific-config> } }
 --
+-- Port names have to be unique by themselves, irrespective of whether
+-- they are free ports or belong to a split-horizon group.
+--
 -- The "config" table contains configuration options specific to a
--- derived class.  It is ignored by the base class.
+-- derived class.  It is ignored by the base class.  A derived class
+-- can access the configuration via self._conf.config.  If config is
+-- not set, it is initialiezed to an empty table.
+--
+-- Note that it is necessary to call the method post_config() after
+-- the app has been configured with engine.configure() to complete the
+-- initialization.  This step will add the ringbuffers associated with
+-- the ports to an internal data structure to save a lookup in the
+-- input and output tables during packet processing.
+--
+-- To make processing in the fast path easier, each port and group is
+-- assigned a unique integer greater than zero to serve as a "handle".
+-- The group handle 0 is assigned to all free ports.
+--
+-- The base constructor creates the following arrays as private
+-- instance variables for efficient access in the push() method (which
+-- must be provided by any derived class).
 --
--- The base constructor checks the configuration and creates the
--- following arrays as private instance variables for efficient access
--- in the push() method (which must be provided by any derived class).
+--  self._ports
 --
---  self._src_ports
+--     Each port is assigned a table containing the following information
 --
---     This array contains the names of all ports connected to the
---     bridge.
+--         { name = <port-name>,
+--           group = <group-handle>,
+--           handle = <handle> }
+--
+--     The tables of all ports is stored in the self._ports table,
+--     which can be indexed by both, the port name as well as the port
+--     handle to access the information for a particular port.
 --
 --  self._dst_ports
 --
---     This table is keyed by the name of an input port and associates
---     it with an array of output ports according to the split-horizon
---     topology.
+--     This is an array which stores an array of egress port handles
+--     for every ingress port handle.  According to the split-horizon
+--     semantics, this includes all port handles except the ingress
+--     handle and all handles that belong to the same group.
 --
 -- The push() method of a derived class should iterate over all source
 -- ports and forward the incoming packets to the associated output
@@ -54,36 +77,111 @@ function bridge:new (arg)
    local conf = arg and config.parse_app_arg(arg) or {}
    assert(conf.ports, self._name..": invalid configuration")
    o._conf = conf
+   if not o._conf.config then
+      o._conf.config = {}
+   end
 
    -- Create a list of forwarding ports for all ports connected to the
    -- bridge, taking split horizon groups into account
-   local ports = {}
+   local ports, groups = {}, {}
    local function add_port(port, group)
       assert(not ports[port],
              self:name()..": duplicate definition of port "..port)
-      ports[port] = group
+      local group_handle = 0
+      if group then
+         local desc = groups[group]
+         if not desc then
+            desc = { name = group, ports = {} }
+            groups[group] = desc
+            table.insert(groups, desc)
+            desc.handle = #groups
+         end
+         group_handle = desc.handle
+      end
+      local desc = { name = port,
+                     group = group_handle }
+      ports[port] = desc
+      table.insert(ports, desc)
+      desc.handle = #ports
+      if group_handle ~= 0 then
+         table.insert(groups[group_handle].ports, desc.handle)
+      end
    end
+
+   -- Add free ports
    for _, port in ipairs(conf.ports) do
-      add_port(port, '')
+      add_port(port)
    end
+
+   -- Add split horizon groups
    if conf.split_horizon_groups then
       for group, ports in pairs(conf.split_horizon_groups) do
          for _, port in ipairs(ports) do
             add_port(port, group)
          end
       end
    end
-   local src_ports, dst_ports = {}, {}
-   for sport, sgroup in pairs(ports) do
-      table.insert(src_ports, sport)
+
+   -- Create list of egress ports for each ingress port, containing
+   -- all free ports as well as all ports from different split-horizon
+   -- groups
+   local dst_ports = {}
+   for sport, sdesc in ipairs(ports) do
       dst_ports[sport] = {}
-      for dport, dgroup in pairs(ports) do
-         if not (sport == dport or (sgroup ~= '' and sgroup == dgroup)) then
+      for dport, ddesc in ipairs(ports) do
+         if not (sport == dport or (sdesc.group ~= 0 and
+                                    sdesc.group == ddesc.group)) then
             table.insert(dst_ports[sport], dport)
          end
       end
    end
-   o._src_ports = src_ports
+   o._groups = groups
+   o._ports = ports
    o._dst_ports = dst_ports
    return o
 end
+
+-- API
+--
+-- Add the ingress and egress links to the port descriptor tables,
+-- accessible via the keys l_in and l_out, respectively.  This helps
+-- to speed up packet forwarding by eliminating a lookup in the input
+-- and output tables.
+function bridge:post_config ()
+   assert(self.input and self.output)
+   for _, port in ipairs(self._ports) do
+      port.l_in = self.input[port.name]
+      port.l_out = self.output[port.name]
+   end
+end
+
+-- API
+--
+-- Print the port configuration and forwarding tables of the bridge.
+-- This is primarily intended for debugging.
+function bridge:info ()
+   local ports, groups = self._ports, self._groups
+   local function nh (n, h)
+      return n.."("..h..")"
+   end
+   print("Free ports:")
+   for p, desc in ipairs(ports) do
+      if desc.group == 0 then
+         print("\t"..nh(desc.name, p))
+      end
+   end
+   print("Split-horizon groups:")
+   for g, desc in ipairs(groups) do
+      print("\t"..nh(desc.name, g)..", members:")
+      for _, p in ipairs(desc.ports) do
+         print("\t\t"..nh(ports[p].name, p))
+      end
+   end
+   print("Forwarding tables:")
+   for p, dst in ipairs(self._dst_ports) do
+      print("\t"..nh(ports[p].name, p))
+      for _, d in ipairs(dst) do
+         print("\t\t"..nh(ports[d].name, d))
+      end
+   end
+end

src/apps/bridge/learning.h

@@ -0,0 +1,68 @@
+/* Type for port and split-horizon group handles */
+typedef uint16_t handle_t;
+
+/* List of egress port handles.  This fixed-sized declaration is used
+   in mac_table.c.  The actual port lists are of variable size and
+   allocated by learning.lua. */
+typedef struct {
+  uint16_t length;
+  handle_t ports[1];
+} port_list_t;
+
+/* Packets are queued in various packet forwarding tables during
+   processing in learning:push().  Each entry points to a list of port
+   handles to which the packet will be sent. */
+typedef struct {
+  struct packet *p;
+  port_list_t *plist;
+} pft_entry_t;
+
+typedef struct pft {
+  uint16_t length;
+  pft_entry_t entries[1];
+} pft_t;
+
+/* Mapping of a destination MAC address to an egress port handle.  The
+   split-horizon group to which it belongs is stored as well in order
+   to detect group collision during lookup.  The address is stored as
+   a 64-bit integer in host-byte order to allow efficient access and
+   comparison.  A bucket in the hash table is made up of an array of
+   BUCKET_SIZE entries of this type. */
+typedef struct {
+  uint64_t mac;
+  handle_t port;
+  handle_t group;
+} mac_entry_t;
+
+/* The MAC addresses are stored in simple hash tables with fixed-sized
+   buckets. */
+typedef struct {
+  uint32_t ubuckets; /* Number of buckets with at least one used slot */
+  uint32_t entries;  /* Number of stored objects */
+  uint8_t  overflow; /* Flag to indicate overflow in at least one bucket */
+} hash_table_header_t;
+
+/* This fixed-sized declaration is used in mac_table.c.  The actual
+   tables are of variable size and allocated by mac_table.lua.  The
+   size of a bucket must be known to access a row in the buckets
+   matrix.  There should be no need to change BUCKET_SIZE for
+   perfomance reasons, see the comments in mac_table.lua. This
+   declaration must match the ctype for hash_table_t in
+   mac_table.lua. */
+enum { BUCKET_SIZE = 6 };
+typedef struct {
+  hash_table_header_t h;
+  mac_entry_t buckets[1][BUCKET_SIZE];
+} hash_table_t;
+
+typedef struct {
+  handle_t port;
+  handle_t group;
+} lookup_result_t;
+
+void mac_table_insert(uint64_t mac, handle_t port, handle_t group,
+                      hash_table_t **tables, uint32_t index);
+lookup_result_t *mac_table_lookup(uint64_t mac, mac_entry_t *bucket);
+void mac_table_lookup_pft(uint64_t mac, mac_entry_t *bucket,
+                          handle_t port, handle_t group, struct packet *p,
+                          pft_t **pfts, port_list_t *flood_pl);