So I've decided to use this blog to provide an running commentary of my thesis in Modeling and Simulation. I'm hoping I stay interested in it enough to update it once every couple of weeks. I know that there are quite a few of topics in this post that I would like to go in-depth on. I'm working on a simulation engine (think game engine with the focus on simulation and not rendering) for massive entity counts such as 100k or higher that scales (in terms of cores and machines) easily and allows for simulation over a planetary scale. First a little background on what I'm doing:
The game engine is an entity bases simulation that uses a component model for the entities. All communication between entities and the components of the system is done through message passing, which allows for much easier multi-threading as well as much easier communication between server back ends. Entities are comprised of behaviors. Behaviors will define an attribute if it is something that needs to be shared with other behaviors or entities. All attributes are cached every frame, so there is temporal consistency during an update phase in a given frame.Tower of Babble:
I had originally thought that most of this engine would be done in C++ and had started working with what I knew, but my advisor insisted on C#. At first I was pissed-I had already done quite a bit of work in C++, which I use for my day job and was very comfortable with. Not to mention I really wanted nothing to do with C#. Give me my trusty C++ (or C for that matter) and let me work with the machine as nature intended it. However, as I started to get into things I found that my productivity was quite a bit higher. It took me significantly less effort to build a capability in C# than it did in C++, and debugging was drastically reduced.Toolbench:
I'm using a number of 3rd party tools to get this running, which I'm sure I'll discuss in quite a bit more detail in the coming weeks. While I was using C++, I used a custom build of GLFW for windowing and input, but since switching to C# I've settled on OpenTK, which is nice since it also has and C# bindings to openAL & openGL. I'm also using ZeroMQ for my message passing infrastructure, which I had recently learned about. ZeroMQ is very cool and pretty fast for what it does. I'll be sure to discuss it in quite a bit more detail later. I use Lua for all in-game scripting as well as any data definition files (such as entity and GUI templates). I'm using LuaInterface for my bindings to C#. I had started using Google's protocol buffers as my message protocol framework, but found that it's lack of inheritance and inability to easily identify a message type from a bytes stream just made it worth the effort to write my own code generator to create message classes off a simple message definition file (again using Lua). I keep going back and forth between Horde3d and Ogre3D for the rendering capability. Again rendering is not the focus, so it's whatever integrates the easiest that will win. I'm leaning towards Ogre3d since there's a bunch of Blender models that I want to use (not to mention I'm a huge Blender fan) and it has a cleaner data pipeline for Blender. Not to mention Ogre3d seems somewhat more active community.LOC-what a stupid metric
I personally think LOC (lines of code) is a horrible metric for estimating effort on a given project, but in this case, it actually is a good metric for measuring the effort between the two similar projects (especially since I was the only engineer working on it). I've finally rebuilt 90% of the existing capability with significantly less code. my original effort was right around 22k lines of code in c++ and my redone effort was about 8k lines of code in C#. I figure another 1k lines to finish that last 10% and have the same functionality with about half the lines of code. That also translates to quite a bit less debugging. My free time is so scarce right now and I'm on such a tight timeline that this kind of productivity boost is a huge win.
Performance tests aren't conclusive yet. My original C++ test application (that did almost nothing) was getting in the ballpark of 2000fps, and the C# is running at around 1400fps on the same machine, but they're not doing exactly the same thing, so it's hard to tell just how much performance is lost to using C# vs C/C++. I'm guessing at this point, as long as it's above 30fps, I'm happy. Let's see what happens when I throw a couple thousand entities at it.
Behold! A spinning box with a cheap crate texture. Let the awesomeness flow.
Kernel and application framework in depth...