EVE Technology Lab

"without any TiDi" is, at this point in time, not really possible. What we do have are dedicated fleet fight nodes, and so long as we are told before downtime that a fight is going to happen via the fleet fight notification on the community web site, we will move that solar system to a fleet fight node all on it's during downtime.

The recent Revenant kill event happened with that solar system on a node all to it's self.

At this point in time there just isn't really hardware available to use to do what we need. The biggest thing to keep in mind is that our server code is not multithreaded, for the most part it all runs on a single core. Processor technology for the last few generations has been real big about expanding out into multiple cores, not higher clock rates. We love higher clock rates. They be good for us.

If they'd spent the money they spent on WoD on re-writing the server code to be multi-threaded they'd be able to support the entire galaxy fighting on one machine* ;)

* Subject to memory bandwidth limitations

Some day CCP will figure out this whole multiple cores thing is not a fad and make the switch I figure about the time 32-64 cores because the norm at the current rate they are going

Didn't say it would be trivial either but since there are already hundreds of other games as complex as Eve and more than that of other types of programs that do it largely correctly every day it's really time that CCP do the same.

The whole 'stack-less' Python idea that is used in Eve is basically the same thing the older AmigaOS did almost 20 years ago and was tried other places but has been replaced by the preemptive way of doing things by every OS still around. Windows, Linux, and OS X all use it and Eve can too Try to do the job of task scheduling in a high level language like Python is never going to out perform the OS working at the machine level and that IS what they are trying to do.
I'll give them credit they've pushed it farther and made it work much longer than most people thought they could but in the end it only works as long as processors keep getting fast and just like CCP FoxFour indirectly pointed out processors really haven't been getting faster clock speeds. This has been true for almost 10 years now and isn't expect to change so multiple cores has been the only way to continue to increase performance in both processors and programs and isn't going to change even if they find ways around the issues currently holding clock rates down like the speed of light, etc.

Okay I'll get off my soapbox now since the thread has kind of got away from the original person's question I have a feel.

Hi all.

Now, i don't normal post in forums, mainly due trolls and flames however occasionally i will post if i see something i feel strongly about.

There appears to be many assumptions in the above statements and a lack of understanding in how eve is set up.

When eve was first begin designed multi-core processors were only just coming out and threading on them was not well supported. For instance, when the first multi-cores came out, it took at least a year before any real programs supported them...

Eve is primarily build on stackless python. This at the time offered many benefits. However, stackless languages work differently from normal languages, eg, they don't use a stack, but rather pass a token around. Changing the eve server to support threading would most likely require a rewrite of the entire server,as threads exist as self contained process on a core passing a token between threads would not be easy, and the overhead for inter-thread communication would be quite big with a stackless setup.

Now, to give you an idea of just how much work that would be, let me refer to good old windows 98. Windows 98 is very old, and a lot smaller than a lot of software out today. and yet...for one man to write windows 98 ..it would take him......168 years...this statistic is from Microsoft its self based on the man hours used to create 98. This should give you a good idea of the scale of rewriting the eve server and clients.

The eve server does in fact use cores, but primarily on the solar system blades. Each on a blade processor 1 core == a node. With this set up, the solar system blades would not be able to thread anyway. And although i know other services like Dogma reside on different servers, you can start see what a huge job it would be to rewrite eve to thread.

I believe ccp has been rewriting bits of its code, in fact Carbon I/O is one such rewrite. I also know from keynotes that they are trying to(or woudl like to) move at least some of the code to c libs. However, considering the above it will take quite a while if ever to thread eve.

In the end, ccp must balance its resources between content and rewrites, Right now there is a huge change in game mechanics, so that new content can be rolled out aka destroy-able everything and player build star gates etc etc.

I would think if ccp decided to thread eve, you would not see any new content or changes for at least 1 year, maybe 2...and i am fairly sure that the stagnation of the game would drive many away, and then whats the point in threading a dead game.

In summary. Eve Could be threaded, but the time and resource to do it in the near future would cripple development of the game features itself, and in the near to medium term is not really feasible.


And before any trolls kick in, Yes i know what i am talking about in regards to threading, but i have dumbed it down for general consumption
Smallevils

If there aren't enough resources available, it's just not possible to do such an overhaul. It doesn't matter what will be. Like the old saying goes: if it ain't broke, don't fix it. Unless eve grinds to a halt, I doubt any serious work will take place. More like workarounds in problematic places.

It depends on what you consider "broken". From a business perspective, it usually looks like this:

- Is it up and brings money?
- Yes.
- It's not broken!

So, unless someone manages to convince upper management to make this investment, Eve will never be brought technologically up-to-date. That's how the IT world works.



You have no idea how much such project would cost. This would most likely be measured in millions.

Well thats a real nice discussion and I am now going to shut up about things I don't understand so well.

Yes, but it was designed this way nearly 15 years ago, when multi-core and multi-threading were going to be released Soon(tm). I don't remember the latest dev blogs on Dogma's design mentioning this, but about the only workaround I realistically see happening is to place each simulation grid in its own thread. Of course, this would break things like probing for PVP targets since one simulation thread would be independent of another unless you do things like cache sharing (which isn't guaranteed to return correct results because the cache could be stale), and I'm pretty sure everybody would agree that's a Bad Thing(tm).

My noobish opinion about how it could be:

Each ship is a separate thread (or possibly something like an Erlang 'process'). When the ship need to interact with another, it sends them a message, something like 'I shoot you with GM Ultra Mega Super Death Ray for 100500 damage', and receives a reply 'you made 1005 damage' (because the tarhet ship has 99% resistances).

Scanning will posibly be something like a broadcast message 'Report all who are in range of 14 au from my position', and the result is displayed after receiving all responces. This can be done asynchronously and doesn't require to pause an entire thread until all other threads respond.

Possibly this will also require a separate physics thread for each active grid.