Routers are basic ingredients of most distributed systems. They commonly have no a priori knowledge about the wider network topology in which they are embedded. Nevertheless, they are expected to send messages via a network port which will move those messages closer to their destinations. Many constraints can be taken into consideration for a practical router, yet this assignment focuses on sending messages on the shortest paths to their destinations.
What an individual router needs to provide to clients:
- Accept a message from a client (in a predefined format). The message also contains a destination id. The expectation is that this message will eventually end at the router of this id.
- Store and provide received messages for a client to pick up (in a predefined format).
What is accessible to an individual router:
- The local network ports which are connected to other routers.
- The ids of all neighbouring routers. In the physical world, this would need to be established by means of initial communications first, yet we assume here that this already happened.
- The total number of routers and the fact that they are numbered consecutively from 1 to the total number of routers. In a real router, you would use hash tables for the router ids.
What the routers need to agree upon before they are deployed:
- One or multiple message formats which are exchanged between routers (in addition to the already given formats used to send to and received from clients).
- The semantics for those messages.
What a router can do:
- Send messages in any of the globally defined inter-router formats over any of its ports to any of its directly connected neighbours.
- Receive messages on any local port and process their contents.
A router may not sneak “behind the scenes” and extract the global topology of the network from a central place in a direct way. Routers can still organize themselves in whichever form they see fit. Thus a central place could, for instance, be one router which by election or other means nominated itself as a central place. Communication with this router would need to go through the normal network channels, though – no secret direct wires to a central coordinator. There does not need to be any central instance at all – distributed solutions are usually also more robust.
You can implement this router in any language or environment, as long as it roughly follows the framework described above. One essential feature for your implementation needs to be kept in mind: the connections between routers are “wires” – not software buffers. A close software equivalent to a wire is synchronous message passing. Any buffer which might be needed must be implemented as part of the router. You can emulate synchronous (or otherwise “immediate”) communication by means of multiple asynchronous messages or asynchronous messages combined with signals/interrupts.
There is a framework available to you in Ada which generates a number of classical network topologies and places router tasks inside a chosen topology. It also runs a simple evaluation that sends messages from each router to every other one and takes notes about the number of hops the message passed through. If you decide to use another language than Ada, please discuss this first with your tutor. Frameworks are available for other languages (see below); if you decide to use a language for which a framework is not provided, then you will need to program this simple infrastructure as well.
The program test_routers can be fully controlled via command line parameters:
accepted options:
[-t {Topology : String }] -> CUBE_CONNECTED_CYCLES
by Size : Line, Ring, Star, Fully_Connected
by Degree, Depths : Tree
by Dimension, Size : Mesh, Torus
by Dimension : Hypercube, Cube_Connected_Cycles,
Butterfly, Wrap_Around_Butterfly
[-s {Size : Positive }] -> 20
[-g {Degree : Positive }] -> 3
[-p {Depths : Positive }] -> 4
[-d {Dimension : Positive }] -> 3
[-c {Print connections : Boolean }] -> TRUE
[-i {Print distances : Boolean }] -> TRUE
[-w {Routers settle time : Seconds }] -> 0.10
[-o {Comms timeout : Seconds }] -> 0.10
[-m {Test mode : String }] -> ONE_TO_ALL
Available modes: One_to_All, All_to_One
[-x {Dropouts : Natural }] -> 0
[-r {Repeats : Positive }] -> 100
Available network topologies are denoted by name (selected from the list above) together with the parameters which are required for a given topology. Thus you can test your routers in a 4-d hypercube by:
./test_routers -t Hypercube -d 4
You can allow your routers some time to get themselves organized and specify for instance:
./test_routers -t Hypercube -d 4 -w 0.2
for a 200 ms preparation phase during which no client messages will be sent through the network.
After this settling time for the routers, the evaluation phase starts, and a message will be sent via every router to every other router. Those messages will be sent in groups, i.e. all messages originating from a specific router will be sent in bulk (“One to all”). This will be changed via the command line parameter -m All_To_One to a schema, where all messages which are destined to a specific router will be sent in bulk. The default timeout for any network activity is 100 ms (but can also be adjusted via the command line interface). If a router does not respond after this timeout, then the communication test is noted as unsuccessful. From the testing environment, it cannot be distinguished whether the message was not delivered or the destination router did not respond, as the internal states of the routers are not accessed. Next, the results of the communication tests are displayed, for example:
----------------------- Instantiating router tasks -----------------------------
=> Routers up and running
-------------------------------- Waiting ---------------------------------------
Time for routers to establish their strategies : 0.10 second(s)
------------------------------ Measurements ------------------------------------
Number of runs : 100
Minimal hops : 1
Maximal hops : 6
Average hops : 3.22
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
+------------------------------------------------------------------------+
1 | . 2 2 3 2 3 4 3 4 3 3 2 4 4 3 5 4 4 6 5 5|
2 | . 2 3 3 2 2 3 4 4 3 3 2 4 5 4 4 4 3 5 6 5|
3 | . 2 3 3 3 2 3 4 4 5 2 2 3 4 4 4 3 4 5 5 6|
4 | 2 2 . 4 3 4 3 2 3 4 4 3 3 3 2 6 5 5 5 4 4|
5 | 2 3 3 . 3 4 4 2 2 4 5 4 3 3 2 5 6 5 4 4 3|
6 | 2 3 3 . 4 4 5 3 2 3 3 4 4 2 2 5 5 6 4 3 4|
7 | 3 2 3 4 3 4 . 2 2 5 4 4 6 5 5 3 3 2 4 4 3|
8 | 2 2 3 4 4 . 2 3 3 4 4 3 5 6 5 3 3 2 4 5 4|
9 | 3 2 3 4 4 5 . 2 3 3 4 3 4 5 5 6 2 2 3 4 4|
10 | 4 3 4 3 2 3 2 2 . 6 5 5 5 4 4 4 4 3 3 3 2|
11 | 3 4 4 2 2 2 3 3 . 5 6 5 4 4 3 4 5 4 3 3 2|
12 | 4 4 5 3 2 3 2 3 3 . 5 5 6 4 3 4 3 4 4 2 2 |
13 | 3 3 2 4 4 3 5 4 4 6 5 5 . 2 2 3 2 3 4 3 4|
14 | 3 3 2 4 5 4 4 4 3 5 6 5 . 2 3 3 2 2 3 4 4|
15 | 2 2 3 4 4 4 3 4 5 5 6 . 2 3 3 3 2 3 4 4 5|
16 | 4 4 3 3 3 2 6 5 5 5 4 4 2 2 . 4 3 4 3 2 3|
17 | 4 5 4 3 3 2 5 6 5 4 4 3 2 3 3 . 3 4 4 2 2|
18 | 3 4 4 2 2 5 5 6 4 3 4 2 3 3 . 4 4 5 3 2 3|
19 | 5 4 4 6 5 5 3 3 2 4 4 3 3 2 3 4 3 4 . 2 2|
20 | 4 4 3 5 6 5 3 3 2 4 5 4 2 2 3 4 4 . 2 3 3|
21 | 4 3 4 5 5 6 2 2 3 4 4 3 2 3 4 4 5 . 2 3 3|
22 | 6 5 5 5 4 4 4 4 3 3 3 2 4 3 4 3 2 3 2 2 . |
23 | 5 6 5 4 4 3 4 5 4 3 3 2 3 4 4 2 2 2 3 3 . |
24 | 5 5 6 4 3 4 3 4 4 2 2 4 4 5 3 2 3 2 3 3 .|
+------------------------------------------------------------------------+
In the provided framework, the routers do not exist yet; thus, this section will not appear up on your screen until you provide a router implementation.
The minimum number of hops for a message should be ‘1’, and the largest measured number of hops should represent the diameter1 of the chosen topology. The average number of hops is characteristic of a specific topology. By using the command line option -i True, you will also see the full table of individual measurements between all nodes. As we only consider bidirectional connections, the number of hops should be identical in both directions between any two routers. If the measurements are different, warnings will be displayed. 1-hop connections are left blank to make the matrix somewhat more readable.
The last section that will appear in the console gives general parameters of the current topology, like the minimum and maximal connection degree and the total number of nodes. You can also print the connections inside the topology as a matrix, which should be the mirror image of the empty spots in the measurement matrix. Double arrows <-> denote bidirectional connections.
--------------- Information about the selected network topology ----------------
Topology : WRAP_AROUND_BUTTERFLY
Dimension : 3
Number of nodes in topology : 24
Constant connection degree : 4
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
+------------------------------------------------------------------------+
1 | . <-><-> <-> <-> |
2 |<-> . <-><-> <-> |
3 |<-><-> . <-> <-> |
4 | <-> . <-><-> <-> |
5 |<-> <-> . <-> <-> |
6 | <-><-> . <-> <-> |
7 | . <-><-> <-> <-> |
8 | <-> <-> . <-><-> |
9 | <-> <-><-> . <-> |
10 | <-> . <-><-> <->|
11 | <-><-> <-> . <-> |
12 | <-> <-><-> . <-> |
13 | <-> . <-><-> <-> |
14 | <-> . <-><-> <-> |
15 |<-> <-><-> . <-> |
16 | <-> <-> . <-><-> |
17 | <-> <-> . <-> <->|
18 | <-> <-><-> . <-> |
19 | <-> . <-><-> <-> |
20 | <-> <-> . <-><-> |
21 | <-> <-> <-><-> . |
22 | <-> <-> . <-><->|
23 | <-><-> <-> . <->|
24 | <-> <-> <-><-> . |
+------------------------------------------------------------------------+
The individual routers need to respond to the following four pre-defined entries:
task type Router_Task (Task_Id : Router_Range := Draw_Id) is
entry Configure (Links : Ids_To_Links);
entry Send_Message (Message : Messages_Client);
entry Receive_Message (Message : out Messages_Mailbox);
entry Shutdown;
Configure is only called once right after the task has been created. The existing links to neighbouring routers are delivered here in the form of an array over the router ids in which connected routers will appear while all other entries are null.
type Ids_To_Links is array (Router_Range) of Router_Task_P;
type Router_Task_P is access all Router_Task;
Two message formats are already defined, yet you need to add at least one more message format and entry that you will use for communications between routers.
type Messages_Client is record
Destination : Router_Range;
The_Message : The_Core_Message;
end record;
type Messages_Mailbox is record
Sender : Router_Range;
The_Message : The_Core_Message;
Hop_Counter : Natural := 0;
end record;
The purpose of Shutdown (which will be called upon once after all tests have been performed) is to allow for a graceful way to stop operations. If your router accepts this call, it is assumed that the router task (and all tasks or resources initiated by it) will terminate eventually. If the call to Shutdown is not accepted, then an abort of this task is attempted by the environment. All this sounds complicated, but it really isn’t. The core of the assignment is to decide in which direction to forward a specific message and to use your knowledge from the course to implement your strategy without locking up unnecessary busy-waiting loops or losing information. In short: your solution should be implementing a router that is responsive on all ports and directs traffic onto the optimal routes.
The basic requirements for implementation in another language are:
- Option to select between and parametrize topologies as provided in the framework.
- Make sure that your routers do not access any information other than the initial connection list to neighbouring routers and the information which they gain by exchanging messages.
- Instantiate the number of routers required to populate the topology.
- The routers need to be instantiated as processes, threads, tasks, or whatever your operating system or language will provide as concurrent entities.
- Provide each router entity with communication links to neighbouring routers. Those links shall resemble a “wire” semantic, i.e. synchronous forms of communication.
- Run a simple test by sending messages from every router to every other router and measure the actual number of hops the messages took.
- Summarize the results as the maximum and the average number of measured hops.
If this sounds too demanding for your chosen environment, then contact us, and we will work out a plan which can be done; a concurrent implementation of the router tasks is obviously mandatory.
Candidates include Go, Rust (basic templates provided for both), Erlang, Smalltalk, LabVIEW, Occam, OpenMP, … just as long as you can rely on message passing (synchronous message passing might be emulated by asynchronous message passing) and support for concurrency.
If we cannot reproduce your environment to run your program, we will potentially ask you to provide more detailed documentation of your tests.
If you have been attentive, you will have noticed a command-line parameter (-x) that controls the number of routers dropping out. After the initial configuration phase (and before normal traffic starts), a specific number of random routers can be powered down purposefully. Many network topologies offer multiple routes to the same node, so the resulting network may still be fully operational (minus any direct communication to the powered down nodes, obviously).
In addition to the basic requirements, the high-distinction-level challenge for this assignment is to design routers that are robust with respect to such losses in the network and will still deliver the packets to the destination, even if the original/optimal path becomes unavailable. In the provided Ada framework, routers will detect the loss of an immediate neighbour by receiving a Tasking_Error exception when trying to communicate to an already terminated router task. Handle this exception and provide an alternative route.
Reflect first for a moment on the following issues (all of which will be related to specific topologies):
- How many dropped out routers can you handle and still guarantee delivery?
- How does the number of dropped out routers affect the performance? You may want to distinguish the situation when you first detect the failure from the following routing operations.
- Can you respond locally, or will you need to propagate the detection of the failure?