Server LOAD test

KEMBL

Hello all, I was make simple load test of trial Photon server "ExitGames-Photon-Server-SDK_v2-0-0-RC1" and glad to publish results for community compare.

Server: Windows 2003 Standard x64 SP2 CPU: 2 x XEON 1.6, MB: Intel VSA5000, RAM: 2 GB RAM, Net 100 Mbit LAN
Client1: Windows Vista Business SP2 / Core2Duo / Net 70 Mbit VDSL, it runs 37 clients
Client2: Windows 2003 Standard x64 SP2 / Core2Duo / Net 100 Mbit LAN, it runs 30 clients
Client3: Windows 2003 Standard x64 SP2 / Core2Duo / Net 100 Mbit LAN, it runs 30 clients

Every client are "Photon-DotNet_v5-7-1-RC2_SDK\demo-mmo\demo-client\Photon.Modules.Mmo.Client.GridDemo" and makes 25 position changes per one second.

Result pictures of load you can see in attach for this post. Exclusive attention needs for Network load graph (green graph in the bottom of every picture). Pictures was make on 70, 80, 90, 97 clients. All clients saw each other, this is most difficult for any game server situation.

All looks good up to 75-80 clients, after this network graph will flow little unstable, updates in client window was little unstable. Up to 90 clients, all was not too bad (satisfactorily). Suddenly, on 97 clients, all stay completely unstable, and test was finished.

During the test, all CPU was evenly loaded, at 97 clients CPU was 68%, Photon server uses ~200Mb RAM. Network utilisation 45 MBit/sec ~ 5 Mbytes/sec.

Conclusion: This is very good results for given test conditions, thank you devs!

KEMBL

One more attach, 70 clients server stats with open client window to show how clients can see each other.

dreamora

Thanks for sharing the data

Depending on what you do, especially when doing an MMO, the graph will look much better as MMOs don't send their positions at a frequency of 25 / second. 10 - 15 times per second and even that often not in form of full transform syncs (I would recommend a bit research on how WoW does that as it uses one of the more interesting approaches and as it there is a self written emulator there was a lot of research on the matter of what it does and its documented on the web)

KEMBL

This was only test with some heavy conditions, of cause, the most of exists ancient style (sword+staf+shield+etc.) MMORPG usually do not need even 5 position changes per second. It is clear for me, cause I was work with it emulators (Lineage2, ragnarok, wow, etc.) for several last years.

The situation changes when you try implement some real MMO evens like siege, or try to build gameplay whole based on fast events flow, like your tank game. In this case suddenly needs to think about in/out traffic, network optimisation/latency and so on. And I think, that to solve such problems is the next step in MMO servers development.

dreamora

The tank game though is no MMO its a normal dedicated server style game.

I can't see any realistic chance to do an MMO FPS action oriented game without growing money trees and wait another 5 years till the FPS environment has been drowned by Ubisoft and EA that far that FPS players are used to pay to play.

Tabula Rasa that was not even real FPS action oriented, yet failed misserably and lost millions. Its known that its development has cost more then 3.5M USD
PlanetSide has a small loyal fanbase but is dead otherwise
Global Agenda thats kind of a GW alike MMO FPS by MMO success scale seems to be a misserable fail too. The amount of players is a joke and it has dropped from 50 EUR to 30 EUR within a month post launch more or less.

I've some doubts that targeting anything like that is even remotely usefull. The vast amount of users is not ready for fast action MMO games or even shooter style games.
Shooter players want their action and want it to be focused -> MMO has little chance, especially if you require a subscription, then its chance to survive goes towards 0
BattleField BC2 already is earning a lot of hate and ignorance due to the lack of dedicated server and the requirement that you must rent it with one of their partners and thats nowhere as "bad" as an MMO with subscriptions or alike.


The only kind of game that I could forsee to benefit of more fast action paced is a GW style game with real action combat or a "Secret of Mana" style MMO.
But even there, the low latency requirement is discussable as the attacks take 0.5 - 1 of time to finish, so even with an 800ms latency it would be no problem. The same holds for most action games actually.

And due to the prediction related suspending of action -> apply action, pings from server to client up to 100ms don't exist at all as thats at least in my code the normal "push back" timeframe for incomming changes against which my own input is applied too.

As you are using Unity give the network example a go. it has very solid base code in relation to that.

KEMBL

I was test Unity3D network already, RakNet is not bad, but I think that Photon will be not less smarter solution. You right, when fall in to some scepticism with FPS based on MMO ideology. The game world have several almost dead solution which tries to make the MMOFPS. The reasons is not only in young technology, or slow Internet connection, or money, or something like that, the main reason is: unknown "Principles of Wheel".

I think this situation usually occurs when all approach seems to be known, and all parts already discovered but main goal still unreachable. The same as Artificial Intelligence, which have all distinct parts discovered (at least in theory), but in real life we have not to smart AI.

Also, by my opinion, the second best way to fall for huge projects is the "ALL IN ONE" idea. They give to gamer so much complex world, that gamer simple have no chance to understand by his whole life, cause the game have got hundreds of Interface options, skills, spells, modes ant etc.

My point of view is that: I definitely can make mmo in classical style, but have no interest to do this. My interest is try to make little bit different kind of game play with simple rules, fast game play and realistic physics provided by unity3D, which now, with out of any optimisation, requires the fat message exchange. So I was test Photon server on high traffic and now I know what this instrument are.

dreamora

I generally agree with you
There are ideas of where to go where the cpu power / costs to achieve it are just out of scale at the time and alike (more sophisticated AI and alike.

Though one thing I would definitely recommend to not assume: That flooding the user, having an overwhelming UI etc will help your game.

It will potentially attract the hardcore player but MMOs life from the less hardcore users, as especially WoW but also the facebook MMOs show.
What makes much more sense is having a complex world with a lot going on, covered behind an intuitive and "growing" environment.
I was back then really positively surprised how Blizzard approach the complexity of their system by continously adding stuff over the course of leveling up to level 10.

Quietus

dreamora wrote:

Thanks for sharing the data

Depending on what you do, especially when doing an MMO, the graph will look much better as MMOs don't send their positions at a frequency of 25 / second. 10 - 15 times per second and even that often not in form of full transform syncs (I would recommend a bit research on how WoW does that as it uses one of the more interesting approaches and as it there is a self written emulator there was a lot of research on the matter of what it does and its documented on the web)

Wow appears to only send input state deltas along. Each client's positional data is trusted. So it's somewhat faked WASD movement, still trying to save bandwidth akin to the old point-click-move method.

In terms of MMO FPS, iirc Planetside circa 2004 limited each side in the 'MMO' to 80 per team per shard. It still felt like a lot of folks when they were all in mech suits, tanks and aircraft of various sort. Certainly more bandwidth restricted in terms of what you can accomplish.

jennifer

Thanks for sharing your work with us.