OS, tools, programming languages, how did the different parts of the software (such as, again, continents) communicate between themselves... For example, I was told once dungeons and raids were/are scripted in Lua.
OS, tools, programming languages, how did the different parts of the software (such as, again, continents) communicate between themselves... For example, I was told once dungeons and raids were/are scripted in Lua.
On launch there was 2 continents, but 4 server blades per server.
Scripted events take place in Lua. But most raids until I believe Cataclysm weren't scripted so much as it was just mob abilities + cool downs.
Note: Not a wow server architect just played way to much WoW
What I mean by this is that currently the only way to have lots of players playing at once in the same realm is to buy very expensive hardware. That plus the amount of DoS attacks you receive if you are popular means that having a private WoW server online is anything but cheap.
I was always more impressed with Asheron's Call management of areas, as they initially had one zone friend world, it would load land blocks as players accessed them.
I've done a lot of reverse engineering of World of Warcraft as I worked on its internals for years. If this is an interesting subject I might write about it. Anything specific people would want to know?
I've mostly learned about the internals of the WoW client itself and how it interacts with the server. I suppose I will eventually write about it when inspiration strikes me, but I'll be happy to answer any more specific questions you have. It's a very broad subject.
>OS, tools, programming languages, how did the different parts of the software communicate between themselves...
I don't know the details of the WoW software architecture but I know there were a lot of weird side-effect bugs in the game.
My favorite weird bug was the one where you couldn't craft a stack of items (like bandages for example) when you were wearing 5 pieces of your class specific items set. So rather than crafting a stack of 20 bandages, you'd have to craft them one bandage at a time if you were wearing 5 pieces of your class set. If you took off one piece of gear you could craft stacks.
I spent a bit of time wondering what type of design would cause that sort of bug....
There's a lot of detail to go into in this field, suffice to say that you can easily make a career out of knowing how to design efficient ESBs.
Obligatory Wikipedia link for more info: http://en.m.wikipedia.org/wiki/Enterprise_service_bus
They said it was more SAN + a centralised Oracle DB. Which is a pretty common approach. The Oracle DB keeps the different blades in sync but also allows trivial data transportation.
XML seems unlikely unless they were storing XML within Oracle but that seems odd/insane for their use case. It might make sense if they were sending data geographically but as far as I know WoW realms are in one location only.
The part I'm talking about is the behind-the-scenes communications layer for the application itself - how you pass information between servers in real time with low latency.
Think of things as a stack. You have the top layer which is the application. It runs on an individual server as a process, or set of processes. You then have middleware, which is software that handles communication between applications, whether it's on the same box, or between servers. Middleware is also responsible for handling communication with the database (oracle in this case possibly). This is the database or information storage layer.
Ie: application layer --> middleware --> database.
This middleware is the portion I'm postulating could be using XML, based on my general enterprise experience. It's not stored anywhere, it's simply a transport mechanism for data between applications when you don't need to store it. More than likely it would contain information such as 'this user is entering the battlegrounds, please create a slot for them with these details', or 'this player just caught the boat to another continent, I'm handing them from me to you and here is all their information'.
XML is great for this kind of 'live' data transportation, although I probably prefer json these days..
What are your savings doing it that way? It just seems to cost more of everything (resources, man-power, etc).
It also helps reduce overhead and latency. Imagine if you had to constantly write the actions of users to a database in real time for 50000 users, including all the metadata, just so your servers could communicate amongst themselves. The latency involved would be huge, even with today's vastly-better equipment. A fast middleware layer greatly helps to cut down on the data that needs to be written to the back end system, while also improving latency, for what's really not a lot of extra complexity. Two birds with one stone, so to speak.
This approach also helps you scale - you can treat everything as a node, with the middleware data as distinct messages. Messages are created and consumed from queues, and so you can have multiple application servers and multiple database servers, with applications publishing messages onto queues for processing, with the database servers consuming these messages if they have the resources to do so. Adding more capacity to the system then becomes as simple as creating a new database server and telling the middleware it exists - the extra resources are automatically used.
Most enterprises work this way.. The basic outcome is that you can start treating the components as a service, with easy separation of duty.
There's more detail here: http://en.m.wikipedia.org/wiki/Middleware
I expect they use a hybrid approach - some monolithic, some message-bus.