DHT simulator

This is an abstract network simulator created for the purpuse of simulating distributed hash tables. It is written in C++17 and depends on the standard library, ssl (for the SHA1 function), and libnop (for serialization, provided in submodule). Either clang or g++ can be used to compile it.

This repository should be cloned with the --recursive flag to also grab the submodules.

$ git clone https://github.com/sid-code/dhtsim --recursive

The network

There are some components to the simulation. Most importantly, there is the "network." In a very abstact sense, it holds all of the participants and transfers messages between them. The network implementation in network.hpp describes one of the simplest possible networks: a fully connected network. Every node can send a message to every other node in the same amount of time.

The network assigns an address to everybody that joins. In the CentralizedNetwork class, that is the template parameter. I usually use uint32_t as the address, but really anything could be used, as most of the classes have the address as a template parameter.

So why is the address type a template parameter? (insert mumblings of eventually using ns-3) I don't know and it was probably a mistake.

The application

Each node is represented by the application that runs on it. See application.hpp for the most abstract definition of an application.

An application is something that can "send" and "receive" messages on the network. At the core, the network can command the application to produce a message to be sent (Application::unqueueOut) or feed a message into it (Application::recv). Every epoch, or "tick", Application::tick is called on every application in the network. Each application then has a chance to do whatever it wants. One can hope that it stored the messages it received in a queue or something like that so that it can access them now.

The message

Just look at message.hpp. That's a message. (The template parameter is the address type again. These things sneak into every piece of code.)

The 'tag' field is to check for responses. If you're responding to a message you got, you use the same tag. That way, the application that sent it knows that it was indeed a response.

The 'hops' field is to count the number of hops the message took. This may be useful later, but I don't use it now.

base.hpp

This file contains a subclass of Application that provides many functionalities such as queues for messages and callbacks for responses.

pingonly.hpp

Finally. This file contains a subclass of BaseApplication that demonstrates how to actually write a real application. This application just responds to pings. See the enum at the top of the file that defines the message types? This is just giving names to some integers.

All this class needs to do is implement handleMessage. The ping method is for convenience. See how it's used by main.cpp.

Sidenote: why is everything in these header files?

Because that's just how C++ template classes work. See [https://stackoverflow.com/questions/1639797/template-issue-causes-linker-error-c](this stackoverflow question) for an explanation. Also, notice the explicit template instantiation in network.cpp. This is to avoid having all the definitions of the network.hpp declarations in the same header file.

What does main.cpp do?

It runs the main loop. Build everything with make and then run the binary it made.