2 // Fraunhofer Institut fuer offene Kommunikationssysteme (FOKUS)
3 // Kompetenzzentrum fuer Satelitenkommunikation (SatCom)
4 // Stefan Bund <g0dil@berlios.de>
6 // This program is free software; you can redistribute it and/or modify
7 // it under the terms of the GNU General Public License as published by
8 // the Free Software Foundation; either version 2 of the License, or
9 // (at your option) any later version.
11 // This program is distributed in the hope that it will be useful,
12 // but WITHOUT ANY WARRANTY; without even the implied warranty of
13 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 // GNU General Public License for more details.
16 // You should have received a copy of the GNU General Public License
17 // along with this program; if not, write to the
18 // Free Software Foundation, Inc.,
19 // 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
21 /** \mainpage libPPI : The Packet Processing Infrastructure
23 The PPI provides an infrastructure to create packet oriented network processing
24 applications. A PPI application is built by combining processing modules in a very flexible
27 \image html scenario.png Target Scenario
29 The PPI concept is built around some key concepts
31 \li The PPI is based on processing \ref packets. It does not handle stream oriented channels.
32 \li The PPI is built around reusable \ref modules. Each module is completely independent.
33 \li Each module has an arbitrary number of \ref connectors, inputs and outputs.
34 \li The modules are connected to each other using flexible \ref connections.
35 \li Data flow throughout the network is governed via flexible automatic or manual \ref
37 \li Modules may register additional external \ref events (file descriptor events or timers).
39 The PPI thereby builds on the facilities provided by the other components of the SENF
40 framework. The target scenario above depicts a diffserv capable UDLR/ULE router including
41 performance optimizations for TCP traffic (PEP). This router is built by combining several
44 \section design Design considerations
46 The PPI interface is designed to be as simple as possible. It provides sane defaults for all
47 configurable parameters to simplify getting started. It also automates all resource
48 management. Especially to simplify resource management, the PPI will take many configuration
49 objects by value. Even though this is not as efficient, it frees the user from most resource
50 management chores. This decision does not affect the runtime performance since it only affects
51 the configuration step.
53 \section packets Packets
55 The PPI processes packets and uses the <a href="@TOPDIR@/Packets/doc/html/index.html">Packet
56 library</a> to handle them. All packets are passed around as generic Packet::ptr's, the PPI
57 does not enforce any packet type restrictions.
59 \section modules Modules
61 A module is represented by a class type. Each module has several components:
63 \li It may have any number of connectors (inputs and outputs)
64 \li Each module declares flow information which details the route packets take within the
65 module. This information does not define how the information is processed, it only tells,
66 where data arriving on some input will be directed at.
67 \li The module might take additional parameters.
68 \li The module might also register additional events.
70 Modules are divided roughly in to two categories: I/O modules provide packet sources and sinks
71 (network connection, writing packets to disk, generating new packets) whereas processing modules
72 process packets internally. In the target scenario, <em>TAP</em>, <em>ASI Out</em>, <em>Raw
73 Socket</em> and in a limited way <em>Generator</em> are I/O modules whereas <em>PEP</em>,
74 <em>DiffServ</em>, <em>DVB Enc</em>, <em>GRE/UDLR</em>, <em>TCP Filter</em> and <em>Stuffer</em>
75 are processing modules. <em>ASI/MPEG</em> and <em>Net</em> are external I/O ports which are
76 integrated via the <em>TAP</em>, <em>ASI Out</em> and <em>Raw Sock</em> modules using external
79 The following example module declares three I/O connectors (see below): <tt>payload</tt>,
80 <tt>stuffing</tt> and <tt>output</tt>. These connectors are defined as <em>public</em> data
81 members so they can be accessed from the outside. This is important as we will see below.
85 : public senf::ppi::module::Module
87 senf::ppi::IntervalTimer timer_;
90 senf::ppi::connector::ActiveInput payload;
91 senf::ppi::connector::ActiveInput stuffing;
92 senf::ppi::connector::ActiveOutput output;
94 RateStuffer(unsigned packetsPerSecond)
95 : timer_(1000u, packetsPerSecond)
97 route(payload, output);
98 route(stuffing, output);
100 registerEvent(&RateStuffer::tick, timer_);
114 On module instantiation, it will declare it's flow information with <tt>route</tt> (which is
115 inherited from <tt>senf::ppi::module::Module</tt>). Then the module registers an interval timer
116 which will fire <tt>packetsPerSecond</tt> times every <tt>1000</tt> milliseconds.
118 The processing of the module is very simple: Whenever a timer tick arrives a packet is sent. If
119 the <tt>payload</tt> input is ready (see throttling below), a payload packet is sent, otherwise
120 a stuffing packet is sent. The module will therefore provide a constant stream of packets at a
121 fixed rate on <tt>output</tt>
123 An example module to generate the stuffing packets could be
126 class CopyPacketGenerator
127 : public senf::ppi::module::Module
130 senf::ppi::connector::PassiveOutput output;
132 CopyPacketGenerator(Packet::ptr template)
133 : template_ (template)
136 output.onRequest(&CopyPacketGenerator::makePacket);
140 Packet::ptr template_;
144 output(template_.clone());
149 This module just produces a copy of a given packet whenever output is requested.
151 \section connectors Connectors
153 Inputs and Outputs can be active and passive. An \e active I/O is <em>activated by the
154 module</em> to send data or to poll for available packets. A \e passive I/O is <em>signaled by
155 the framework</em> to fetch data from the module or to pass data into the module.
157 To send or receive a packet (either actively or after packet reception has been signaled), the
158 module just calls the connector. This allows to generate or process multiple packets in one
159 iteration. However, reading will only succeed, as long as packets are available from the
162 Since a module is free to generate more than a single packet on incoming packet requests, all
163 input connectors incorporate a packet queue. This queue is exposed to the module and allows the
164 module to process packets in batches.
166 \section connections Connections
168 \image html ratestuffer.png Simple RateStuffer
170 To make use of the modules, they have to be instantiated and connections have to be created
171 between the I/O connectors. It is possible to connect any pair of input/output connectors as
172 long as one of them is active and the other is passive.
174 It is possible to connect two active connectors with each other using a special adaptor
175 module. This Module has a passive input and a passive output. It will queue any incoming packets
176 and automatically handle throttling events (see below). This adaptor is automatically added by
177 the connect method if needed.
179 To complete our simplified example: Lets say we have an <tt>ActiveSocketInput</tt> and a
180 <tt>PassiveUdpOutput</tt> module. We can then use our <tt>RateStuffer</tt> module to build an
181 application which will create a fixed-rate UDP stream:
184 RateStuffer rateStuffer (10);
186 senf::Packet::ptr stuffingPacket = senf::Packet::create<...>(...);
187 CopyPacketGenerator generator (stuffingPacket);
189 senf::UDPv4ClientSocketHandle inputSocket (1111);
190 senf::ppi::module::ActiveSocketReader udpInput (inputSocket);
192 senf::UDPv4ClientSocketHandle outputSocket ("2.3.4.5:2222");
193 senf::ppi::module::PassiveSocketWriter udpOutput (outputSocket);
195 senf::ppi::module::PassiveQueue adaptor;
197 senf::ppi::connect(udpInput.output, adaptor.input);
198 senf::ppi::connect(adaptor.output, rateStuffer.payload);
199 adaptor.qdisc(ThresholdQueueing(10,5));
200 senf::ppi::connect(generator.output, rateStuffer.stuffing);
201 senf::ppi::connect(rateStuffer.output, udpOutput.input);
206 First all necessary modules are created. Then the connections between these modules are set
207 up. The buffering on the udpInput <-> rateStuffer adaptor is changed so the queue will begin to
208 throttle only if more than 10 packets are in the queue. The connection will be unthrottled as
209 soon as there are no more than 5 packets left in the queue. This application will read
210 udp-packets coming in on port 1111 and will forward them to port 2222 on host 2.3.4.5 with a
211 fixed rate of 10 packets / second.
213 \section throttling Throttling
215 If a passive connector cannot handle incoming requests, this connector may be \e
216 throttled. Throttling a request will forward a throttle notification to the module connected
217 to that connector. The module then must handle this throttle notification. If automatic
218 throttling is enabled for the module (which is the default), the notification will automatically
219 be forwarded to all dependent connectors as taken from the flow information. For there it will
220 be forwarded to further modules and so on.
222 A throttle notification reaching an I/O module will normally disable the input/output by
223 disabling any external I/O events registered by the module. When the passive connector which
224 originated the notification becomes active again, it creates an unthrottle notification which
225 will be forwarded in the same way. This notification will re-enable any registered I/O events.
227 The above discussion shows, that throttle events are always generated on passive connectors and
228 received on active connectors. To differentiate further, the throttling originating from a
229 passive input is called <em>backward throttling</em> since it is forwarded in the direction \e
230 opposite to the data flow. Backward throttling notifications are sent towards the input
231 modules. On the other hand, the throttling originating from a passive output is called
232 <em>forward throttling</em> since it is forwarded along the \e same direction the data
233 is. Forward throttling notifications are therefore sent towards the output modules.
235 Since throttling a passive input may not disable all further packet delivery immediately, all
236 inputs contains an input queue. In it's default configuration, the queue will send out throttle
237 notifications when it becomes non-empty and unthrottle notifications when it becomes empty
238 again. This automatic behavior may however be disabled. This allows a module to collect incoming
239 packets in it's input queue before processing a bunch of them in one go.
241 \section events Events
243 Modules may register additional events. These external events are very important since they
244 drive the PPI framework. Possible event sources are
245 \li time based events
246 \li file descriptors.
247 \li internal events (e.g. IdleEvent)
249 Here some example code implementing the ActiveSocketInput Module:
252 class ActiveSocketReader
253 : public senf::ppi::module::Module
255 typedef senf::ClientSocketHandle<
256 senf::MakeSocketPolicy< senf::ReadablePolicy,
257 senf::DatagramFramingPolicy > > SocketHandle;
258 SocketHandle socket_;
259 DataParser const & parser_;
260 senf::ppi:IOSignaler event_;
262 static PacketParser<senf::DataPacket> defaultParser_;
265 senf::ppi::connector::ActiveOutput output;
267 // I hestitate taking parser by const & since a const & can be bound to
268 // a temporary even though a const & is all we need. The real implementation
269 // will probably make this a template arg. This simplifies the memory management
270 // from the users pov.
271 ActiveSocketReader(SocketHandle socket,
272 DataParser & parser = ActiveSocketReader::defaultParser_)
275 event_ (socket, senf::ppi::IOSignaler::Read)
277 registerEvent( &ActiveSocketReader::data, event_ );
278 route(event_, output);
287 output(parser_(data));
292 First we declare our own socket handle type which allows us to read packets. The constructor
293 then takes two arguments: A compatible socket and a parser object. This parser object gets
294 passed the packet data as read from the socket (an \c std::string) and returns a
295 senf::Packet::ptr. The \c PacketParser is a simple parser which interprets the data as specified
296 by the template argument.
298 We register an IOSignaler event. This event will be signaled whenever the socket is
299 readable. This event is routed to the output. This routing automates throttling for the socket:
300 Whenever the output receives a throttle notifications, the event will be temporarily disabled.
302 Processing arriving packets happens in the \c data() member: This member simple reads a packet
303 from the socket. It passes this packet to the \c parser_ and sends the generated packet out.
305 \section flows Information Flow
307 The above description conceptually introduces three different flow levels:
309 \li The <em>data flow</em> is, where the packets are flowing. This flow always goes from output
311 \li The <em>execution flow</em> describes the flow of execution from one module to another. This
312 flow always proceeds from active to passive connector.
313 \li The <em>control flow</em> is the flow of throttling notifications. This flow always proceeds
314 \e opposite to the execution flow, from passive to active connector.
316 This is the outside view, from without any module. These flows are set up using
317 senf::ppi::connect() statements.
319 Within a module, the different flow levels are defined differently depending on the type of
322 \li The <em>data flow</em> is defined by how data is processed. The different event and
323 connector callbacks will pass packets around and thereby define the data flow
324 \li Likewise, the <em>execution flow</em> is defined parallel to the data flow (however possible
325 in opposite direction) by how the handler of one connector calls other connectors.
326 \li The <em>control flow</em> is set up using senf::ppi::Module::route statements (as long as
327 automatic throttling is used. Manual throttling defines the control flow within the
328 respective callbacks).
330 In nearly all cases, these flows will be parallel. Therefore it makes sense to define the \c
331 route statement as defining the 'conceptual data flow' since this is also how control messages
332 should flow (sans the direction, which is defined by the connectors active/passive property).
334 \see \ref ppi_implementation \n
335 <a href="http://openfacts.berlios.de/index-en.phtml?title=SENF:_Packet_Processing_Infrastructure">Implementation plan</a>
338 /** \page ppi_implementation Implementation Overview
340 \section processing Data Processing
342 The processing in the PPI is driven by events. Without events <em>nothing will happen</em>. When
343 an event is generated, the called module will probably call one of it's active connectors.
345 Calling an active connector will directly call the handler registered at the connected passive
346 connector. This way the call and data are handed across the connections until an I/O module will
347 finally handle the request (by not calling any other connectors).
349 Throttling is handled in the same way: Throttling a passive connector will call a corresponding
350 (internal) method of the connected active connector. This method will call registered handlers
351 and will analyze the routing information of the module for other (passive) connectors to call
352 and throttle. This will again create a call chain which terminates at the I/O modules. An event
353 which is called to be throttled will disable the event temporarily. Unthrottling works in the
356 This simple structure is complicated by the existence of the input queues. This affects both
357 data forwarding and throttling:
358 \li A data request will only be forwarded, if no data is available in the queue
359 \li The connection will only be throttled when the queue is empty
360 \li Handlers of passive input connectors must be called repeatedly until either the queue is
361 empty or the handler does not take any packets from the queue
364 \section logistics Managing the Data Structures
366 The PPI itself is a singleton. This simplifies many of the interfaces (We do not need to pass
367 the PPI instance). Should it be necessary to have several PPI systems working in parallel
368 (either by registering all events with the same event handler or by utilizing multiple threads),
369 we can still extend the API by adding an optional PPI instance argument.
371 Every module manages a collection of all it's connectors and every connector has a reference to
372 it's containing module. In addition, every connector maintains a collection of all it's routing
375 All this data is initialized via the routing statements. This is, why \e every connector must
376 appear in at least one routing statement: These statements will as a side effect initialize the
377 connector with it's containing module.
379 Since all access to the PPI via the module is via it's base class, unbound member function
380 pointers can be provided as handler arguments: They will automatically be bound to the current
381 instance. This simplifies the PPI usage considerably. The same is true for the connectors: Since
382 they know the containing module, they can explicitly bind unbound member function pointers to
386 \section random_notes Random implementation notes
388 Generation of throttle notifications: Backward throttling notifications are automatically
389 generated (if this is not disabled) whenever the input queue is non-empty \e after the event
390 handler has finished processing. Forward throttling notifications are not generated
391 automatically within the connector. However, the Passive-Passive adaptor will generate
392 Forward-throttling notifications whenever the input queue is empty.
394 \section class_diagram Class Diagram
396 \image html classes.png
403 // c-file-style: "senf"
404 // indent-tabs-mode: nil
405 // ispell-local-dictionary: "american"