|
Author |
Thread Statistics | Show CCP posts - 0 post(s) |
Eldariel
|
Posted - 2004.01.26 10:13:00 -
[1]
I must admit I did laugh when Tank and a bunch of others jumped in the help channel demanding the servers be taken down and more nodes be allocated. Like those 6300 other players on the servers at the time don't matter at all (rofl)
I'm also intrigued how everyone is so sure you can just throw more nodes at this to solve the problem, since it assumes the technical design of the application is infinitely scalable (system can be load balanced across as many nodes needed).
Could just as easily be that the application is only partially scalable - i.e. they can allocate across multiple nodes, but the number is finite. If that's the case there will *always* be a technical limit to size of fleet battles unless there is a *major* rewrite of the server code. Not likely to happen anytime soon ...
In a fleet of 150 ships, with 5-10 spawned items per ship at any moment in time (weapon fire, missiles etc), there are going to be something like 1500 objects in space. To my mind it's hardly surprising that things grind to a halt
Not saying that CCP shouldn't look into it - but don't expect it to go away anytime soon if it's a design issue ...
|
Eldariel
|
Posted - 2004.01.26 10:19:00 -
[2]
Edited by: Eldariel on 26/01/2004 10:25:52
Quote: Why would they design it to be this bad though ?
Surely they realised that people would amass fleet this large ? Surely they realised that 5000+ people would be online a a single time ?
I mean, was a battle of this size not tested in Beta ?
Not saying they have - just that it is possible. Only the devs know for sure..
Having said that I vaguely remember some dev chats last year where it was implied the design wasn't infinitely scalable at a system level - could be just me imagining it though
Worth noting that lag wasn't bad elsewhere - so 5000+ spread over the universe is fine. Just don't expect to see them all in the same system at the same time anytime soon
The point is it's an assumption that throwing more nodes at it will make the problem go away. It may not ..
|
Eldariel
|
Posted - 2004.01.26 11:05:00 -
[3]
Edited by: Eldariel on 26/01/2004 11:06:27
Quote: not infinite scalability, greatest scalibility. I took this to mean that they can dedicate a single node to a single system, rather than one node handling several systems.
Single node per system would be bad - since the number of supported players would be directly linked to the size of this single box.
I've a feeling they might have it scalable at a region or constellation level, but even then there may be a limit to the no. of nodes in the micro-cluster
|
Eldariel
|
Posted - 2004.01.26 15:17:00 -
[4]
Edited by: Eldariel on 26/01/2004 15:21:52
Actually Oracle corp (RL - not in-game ) is heavily pushing the latter approach (i.e. wintel cheap boxes running linux, loads of redundancy, but a lot of them) with grid computing
If correctly designed (from an application and an architecture perspective) a grid of small boxes can theoretically be much better than one big box. Getting it right from the outset is pretty difficult though - quite easy to miss potential bottlenecks during design, which is probably whats happening here
To be fair I've yet to see a MMORPG that didn't have issues with lag and/ or slowdown (whether client or server side) when large numbers of players congregated in a single area or zone. Most just bounce you to a different zone/ area when the load gets too high
Odd though - the fact not all are lagged makes it sounds like a load balancing issue - i.e. not distributing load across all nodes correctly. If so it may be "relatively" easy to fix
|
Eldariel
|
Posted - 2004.01.26 16:21:00 -
[5]
Edited by: Eldariel on 26/01/2004 16:25:35
I agree
That's still a design issue though - the application design should have assumed a 10 mbps LAN transfer rate, not 100 mbps . In effect the LAN has become the bottleneck ...
All architectures have their limitations and pitfalls in one way or another ... the key is balancing these to get optimal results. Optimal results in terms of performance and architecture does not necessarily equate to end user expectations though
|
Eldariel
|
Posted - 2004.01.26 23:34:00 -
[6]
Edited by: Eldariel on 26/01/2004 23:37:57
Quote: Several GMs have told us that if we let them know about a fight before downtime they can boost that node. Only problem is we need the system name and as most battles happen near midnight eve time, that is a 12hr warning we have to give which is just crap
Thats a common technical limitation on most server applications - server initialization parameters (e.g. those allocating memory etc) quite often can't be changed on the fly. Hence they need a reboot - and this, by inference, probably means a cluster reboot if there are parameter hooks
You didn't/ don't *really* expect them to bounce the entire server/ cluster on demand do you? What about the other 6300 players ...
|
|
|
|