Tobold's Blog
Saturday, December 17, 2011
The queue is here

It appears that EA Bioware let in a lot of players yesterday. The result was predictable: While server loads where between "light" and "normal" on the first 3 days of early access, they now went all the way up to "full". I relogged once during the evening to try out an .ini file modification to force anti aliasing, and had to wait in a queue half an hour to get back in. And that wasn't the longest queue.

The problem is that for a given server population, the server load depends on the percentage of players wanting to play on that particular evening. In the long run that is between 10% and 20%, there is an industry rule of thumb that overall server population is between 5 and 10 times prime time population. But during early access and launch this percentage is significantly higher, because pretty much everybody wants to play NOW.

I don't have the technical knowledge to judge why it is still necessary to run big MMORPGs with separate servers instead of one single server distributing players in many instanced carbon copies. But as long we have servers, and we don't want those servers to be underpopulated after launch, we will have those queues on launch day and early access.
Let me echo your disappointment that more games have not adopted the dynamic carbon copy server approach.

For players it solves the problem of queues and the problem of playing with your friends. For the game company is saves them from publicly embarrassing server closures once the initial bubble has passed.

I have hopes that Guild Wars 2 will continue with the server model that worked so well in Guild Wars 1 but I haven't seen absolute confirmation of this.
I would assume it isn't a technical issue at all, but a social one.

Ignoring the whole server identity aspect of it, just imagine the practical effect of what happens when it is impossible for you to ever meet an individual more than once. Even if there were some convoluted system where guilds and "friended" players always appear in the same instance of whatever city hub, everyone else would always and forever be a temporary stranger. It would be the equivalent of LFD, all the time: Looking For Server.
Can you cite a source for that industry rule? When I first heard it, back in around 2002-3 in an interview John Smedley of SoE gave to a trade publication (which I've tried to track down since without success) I remember it being quoted as a one in three ratio - one third of players online for two thirds offline at peak.

I've always used that to work out server populations. Obviously if it's actually 1:5 or even 1:10 I've been vastly underestimating those populations.
"I don't have the technical knowledge to judge why it is still necessary to run big MMORPGs with separate servers instead of one single server distributing players in many instanced carbon copies."

It's probably possible, but there are downsides. Servers form a community with a limited number of players. If servers got split in two at peak time, people would have to cross between instances to interact with half their friends.

Possibly it might be worth it. One option would be that if this happened a lot, players would expect asn eventual perma-split of the server, and could make plans accordingly.

On the other hand, popular servers might end up continually growing and splitting.

It would at least be a novel mechanic, I suppose. Maybe it could even be harnessed as part of the game scenario... some weird SF or magical effect.
While queues are annoying, Bioware needs to be disciplined and avoid the "RIFT error." RIFT was probably the smoothest MMO launch I've ever experienced, but they were unprepared for the sheer number of players they had at launch. They overcompensated, added a TON of additional servers, and six months later half those servers turned into ghost towns without sustainable player populations. This is where I think the phased pre-launch is genius. Bioware can control the number of players initially and use information gathered during early access to predict server load and demand for the actual live launch.

The best chance for healthy servers later is to just accept there will be some queue times for launch and probably a good month or so after. Once the 30-day "tourists" bail out, we'll see where the populations stand. But the players, and Bioware, just need to be patient until then. An overreaction now will have consequences later. Just wait it out.
it works wonderfully in guild wars, yes, but I think part of the reason for that is that GW is barely an MMO. the only parts of it that you say players in are the major cities. every player who goes out for quests gets their own instance. you never run into anyone in the world, unless you specifically group with them so that you go out into the world together. I enjoy guild wars very much when I play it. but in all honestly, it feels more like a massively co-op to me. there is no need for a persistent world.

I have a feeling that guild wars 2 will go with multiple servers. becasue from what I know so far, they plan on creating the world that changes with how players affect it.. and it changes for everyone. and that means that they would probably have to split it into smaller persistent chunks of the world, for those of us whose computers cannot handle a strain of rendering that many people.

but why not keep it with different isntance each time, you see? well, lets say I protected the village from the centaurs before I logged out and I loge back into an instance where no one protected the village.. or helped sack it. kinda ruins immersion (I know its a dirty combination of words, but I don't know how else to say it)
"I don't have the technical knowledge to judge why it is still necessary to run big MMORPGs with separate servers instead of one single server distributing players in many instanced carbon copies."

Champions Online tried this, and as others have mentioned, it made it feel like you were never going to see any of these people again. It was like using the random dungeon finder all the time. How many friendships did you develop using that?

The better solution is from City of Heroes. Every zone has the potential to create another instance. Most of the time, for most zones, there is simply the one instance for the server. But in high population times, another can be created.
Given global lag, you definitely want a separate server per global region. I know an Australian guildie whose ping times went from 450ms to 170ms by moving from an US east coast server to US west coast during beta.

Within a geographic region, you should be able to have a single server with heavy instancing, like Champions Online. One can work around the social shortcomings of instances by making them named and sticky. For example, a user may choose to go to the "Jar Jar Binks" instance and his/her character will remain at that instance until the user changes it. In a way, this is like servers, except the character can move from server to server in game at any time. I don't know why MMOs don't do this.

The only technical reason I can think for multiple servers is database scaling issues. The bottleneck for modern MMOs is probably database IO bandwidth. It is much easier and cheaper to scale up by creating multiple small, independent databases than a single huge, central database.
It strikes me that the phased early access might have it's problems - At least a few people from most of the planned guilds are going to be the earliest on early access servers and will start their guilds on those servers. But on the 20th there will be another rush of players and they will want to join their guilds. In that case the load on those servers can only get worse.

Guild wars 2 is going to adopt an individual world model since they want populations to form communities. I wouldn't be surprised though if they are able to use the technology developed for guild wars 1 to have a virtual server approach so that the connection between the worlds and the physical hardware is not 1-1. In that case they might find it easier to scale up and down.
"just accept there will be some queue times for launch".

Yes, except that those I heard in the game were talking of queues of 2 a 4 hours long.

I want to play for two hours an evening, not wait for two hours.
The problem comes from having characters affixed to a specific server. Character data should be separate and universal. Even the browser-based runescape can do this. If a world is busy, players just choose a low pop world to get in to play together.

The added benefit to this is that after the launch rush is over, servers can shutdown anonymously and avoid character transfers or perceptions that the game is "Phail".
Post a Comment

<< Home
Newer›  ‹Older

  Powered by Blogger   Free Page Rank Tool