I’m trying to find info on a game engine that supposedly failed called Wisp. Some of the aspects of this engine that I’ve heard about have my interest and I’m trying to find a website on it or contact info for the maker of the engine. From what I heard the engine failed to meet the 2.5 million player limit that the creators wanted and they dumped the engine.
Thanks in advance to anyone who knows something about it.
2.5 million players? beware my screen has fewer pixels than this game is supposed to have players!
the thing you’r looking for might be hard to find. failing at such ammount of player (i would guess its some sort of mmo-game) is definetly a server issue. since clients (where the engine usualy is used) only handle the currently visible part of the world where the player is in.
good luck on your search,thought
yeah I was thinking the same thing that it was a server issue or they really screwed up on script code somewhere. From my understanding of the info that was given to me this was a new engine for next generation games and was being used for a game the engine creators were making. They supposedly started the server stress testing on the game and it failed to meet the specs they wanted. My brother’s friend was one of the ones doing the stress test so I have to wait to speak to him to see if he has any information on the engine or it’s creators. One of the things this engine supposedly had was a feature on it’s mobs that allowed them to roam the whole game world and form societies. They would them build towns, cities, forts, and many other things just like the AI controlled players in any RTS type game. My thoughts are first ‘hey what a great idea’ and second ‘hey this and the 2.5 million players together would be a possible overload on the server’
So far my searches on this engine have turned up nothing.
Ok I got to speak in person to one of the stress testers and found out that the engine is not called wisp. There was no name to the engine as it was a custom one being used to make the game Wish. projectwish.com
I am now trying to talk to the Devs of Project Wish.
why are you so eager about getting hands on this engine^ panda is great too. and if you have a server which can handle the 2.5 million players … well panda has no problem with dynamic levels where the user can build houses, or even forts etc. 
No I don’t want their engine. I was after the scripting for the mobs and npcs that allows them to roam around and build towns and stuff at random places. I also wanted to know if it was this that crashed out their server during the test. We don’t want something in our game just because it looks or sounds like a good idea. We want to know if it works and is not going to be problematic later on.
On a side note: We now have an IT specialist on our team that is coming on board with servers, network, backup, software, and many other items in tow to aid in the hardware and some of the software side of our game that isn’t engine related. He has the stuff to test many different client side hardware and software configurations along with many of the new development systems with the software and lic that are equal or better than the ones we are using now.
the code will be engine specific …
most scaling problems for me are solved by amazon ec2 … just finding money to pay for it …
still… try to find a engine that handles 2.5 million users. o0
I am not talking about server issues here. You need a damn good algo to be able to have that ammount of players online at a time. 
Think about the calculations per second… 
Regards, Bigfoot29
A damn good algo… and a damn good game to wrangle that many people!
I don’t know anything about MMORPG, but with the only way I can see it scale is by using an algo similar to these used by cellphone companies to assign cell towers to incoming transmissions (on a large buncha o’ servers ).
there are many ways of handling tons of players. one very intresting amongst them is to use a p2p based architecture
this way allows to share most of the nececssary processing and networking amonst the clients themselves. it’s quite complicated, but in such an extreme decentralized system the server would only handle incomming connections and integrate them into the existing network, if existing at all.
security is quite an issue with this network. same with presistence of the game world, since there is no server which could be querried for the “true and ultimate state of the gameworld”.
you can imagine it as an pseudo-democratic-anarchic game with self-evolving capability. where the game rules itself based on majority and self-judging.
one nice thing about it: it’s self-scaling and there are no gameservers which could crash. aside from that it’s horrible to maintain cause it lacks consistence.
when it comes to a database which stores client accounts etc. one either has to use a permanent server providing this information. or use something similar to darknet.
if anyone is interested in how such a freaking thing would work without breaking i could try to make some concepts of working for it.
I’ve heard EVE online uses stackless python. (It shouldn’t be too hard to make panda work with stackless.)
I am interested in how a p2p architecture could be made to work.
They do – I think there are some articles on Gamasutra I believe about EVE & Stackless Python. I’m pretty sure much of it by this time is in-house hacked.
Also, they did run into performance issues with Stackless dealing with number of players. They did some big redesigning of things a year or so ago because they were seeing fights much larger than they anticipated – apparently around a hundred players in the same fight causing Stackless to slow to a crawl. Not sure what they did to fix this but they released some presentation video of their developers explaining this and showing the before and after demonstrations.
That was for gamers not developers so it wasn’t super technical but did give some insight.
(Note: I don’t play EVE so the above is just from keeping up with MMOs not from a player’s POV)
well about the p2p based thing. its sorta tricky but i threw together an approach which is still somewhat comprehensive.
Cyan: hee… well now you did it 
make sure to read everything… or else you wont get dinner
first off: to keep things simple lets keep the gameworld mostly flat so we can work with 2 dimensions. more are possible but make things unneccssary complicated. you could reduce the dimensions to 1 , which would help a lot in understanding. i might do so later on.
the most basic idea is that every client is a server at the same time, and only deals with other clients/servers which are close to him (in the sense of “their avatars are close to each other in the game world”) so it performs a few tasks of sharing and redirecting data.
that’s all there is… but now for the more interesting part of the idea
so lets assume we have a level. first we do is divide it into spaces. an easy way is to split horizontally then vertically , horizontally , and so on. this would be called binary space partitioning. so we can roughly tell where the player is located based on which side of each split he is. there number of splits depends on mapsize. it doesnt matter if you split a few times too often.
well now take the smallest area you got from this slice. lets call it “area” for convenience.
well now we have players. a player always is in one specific area. that’s ok. but not very suited for later purpose. so lets extend the region in which a player belongs to an “area”. make it twice as big, center at the same spot. lets call this area “sharingArea”. so a player is usualy in 4 “sharingAreas” at the same time.
now… more preperations. every client, is not only a client. but also a server.
to keep things simple we use an login server to let the clients meet each other once.
after than its getting tricky. first we need a “netsplit” value which decides how many players can co-exist in a region before they are re-grouped into smaller groups.
and even more preperations. each area has its own “server” , which is basically only a client handling a at most the number of clients set by netsplit.
other clients can send data to this one, and he will share it amongst the clients in his area. if he has more clients then specified by the netsplit value, the clients will get grouped by the area they are in. one of the clients of the old group will become a “groupserver”. the old groupserver will also send his data to the newly create child-groupserver. this groupserver will share data again.if he getes some.
now more tricky stuff. if a groupserver gets a package. he will redirect to all his clients which are on the “closer” side of the map-splitting thing. (its easy to dertermine based on the area-code and avoids that data gets shared all over ) after redirecting he will send a message back to the original sender that his package was send to him, with his own address. so the original sender can send his next data directly to the groupserver in charge for the closest neighburing area.
to understand this it’s neccessary to understand the sending machanism. i’ll try to explain with an example.
this is quite easy, there always is at least one group per game, even if only one player is present. the clients send a request to the groupserver, the groupserver then returns 2 lists with adresses, the one contains all clients which are in the current group, so to say, the list where the data is send directly without beeing redirected. and a list with groupservers which would typically be the closest servers around on the groupserver (logic wise). each address comes with the area code belonging to it. groupserver addresses have a “deepth”" so their area code is missing the most “detailed” part. it only contains the area code up to the place where they are responsible for managing the data.
the mechanism works like this:
first client connects. joins the world, the login server keeps his address data.
the client is all alone, he becomes a “areaServer”, calculating what’s his “area” , and all “sharingAreas” he’s in. then he checks if other clients are connected he could share his data with.
nothing to do.
then a second cliend connects. the loginserver gives the address from clien1 to client2.
client1 and client2 connect to each other.both calculate their area and sharing area. sending data to each other.both check each others “area” value and check if they have childGroups. no child groups present. everything is clean and simple
more clients connect. and at some point a first netsplit happens. a subgroup is created.
now lets say client9 is the groupserver of the subgroub.
client 5,6,7,8,9 are in the subgroup and 1,2,3,4,9 in the original group. as groupserver the client9 is defacto part of the subgroup AND the groub at the same time.
so if client 1 now sends data to his group-members, client9 will receive it, and send it to his clients which are closest to client1 … all based on the area codes.
more stuff is waiting. if now a client of a subgroup (which is nothing else than a group with a parentgroup, code is the same) has data to send, he sends it to his group members. he also will send his data to the group-server. the groupserver will check if there are neighburing groupservers he knows about, or if the closest neighbors are already part of his group. if he cant find a neighboring groupserver, he will redirect the data to his groupserver again, if one is present. this one will do the same. until either the top is reached, or a neighboring groupserver is found. if one is found the data is send to him, this one will then send the data to his closes childs again, and a message back to the oringinal sender to let him know he is the closet match found for the path the data took until now. now if there are groupservers amongst the receivers of the server which was found as neighburing groupservers, they will again send their data to the closest neighbor and again a message back to the original sender.
this continues until the last branch was reached.
this can cause the first package to take quite a detour. but since the client got noticed about who is his closest neighbor groupserver. the next package from him can be send directly to him, which then gets distributed, so during normal operation the data makes only a single “hop”.
since it can happen that a groupserver disconnected , the process will continue once again do search for closest neighbors.
in case a groupserver just moved away so it’s not longer the “closest” match, he will just notify the client he got some data from and tell him to find someone else in his case.
if a group loses members for what reason ever, groupservers can cancel the group, and merge their clients with the topgroup.
well the whole thing is tricky, but still reasonable small to understand it. your application should be very timing-uncritical. the whole thing can be used to share things like movement data, player properties such as outfit, inventory, stats etc.
presistent data is not meant to be shared with this system. so player accounts, items lying around etc are not part of the whole thing. same with npc’s, which could run as pseudo-clients in the same network.
if a system for storing data such as accounts etc is used, clients can start to monitor each other actions to prevent cheating in a limited manner.
since clients can see what other clients are doing they can “check” on their actions. so things like … the lvl of a player jumped without a reason, the other clients can “suspect” him and log the actions in question. is suspect is grave, the logs can be send to a gamemaster or admin which can manually check the data if this is really bad, and take further steps.
if you leave away the human judging and just check if other clients data are conform to the own simulation you can get a self-evolving system if enough people hack their own client.
well enough for now. if you are still intrested in such an brainmelting idea lemme know. i’ll try to somehow visualize the thing so one can actually imagine what’s going on.
now that it’s 1am, and now that i made at least a dozens of gramatical errors. i’ll go to bed and have some rest^ thx for reading, in case you did i wouldn’t have done it if i where you
Hmm, I’m not sure how well I followed your explanation. So basically you divide your players into a tree? The original server becomes the root, and the group servers become internal nodes. The leaf nodes act as clients but not servers. The netsplit number is the maximum branching value, and the division is by area? In the game’s space?
well thats how it works very well summarized.
Why branch by gamespace? It would seem to encourage cheating, since a hacked program could potentially affect the performance of the corresponding player.
It seems to me that the branching should be by latency time instead of gamespace. This would improve performance of the of the system as a whole, and it would reduce the motivation for cheating. The area the server module serves should change randomly and be far away (in gamespace) from the area the client module sees on the same machine. If your hacked client can’t affect the area your player is in (or even reliably affect any particular location or player), then it can’t help you and what’s the point?
such systems are always open to hacking attacks in general. grouping by network performance CAN improve network performance. but makes managing the whole thing more difficult. determining who sends the data to whom is not easy this way.