History
Games tend to use the chess rating systems as a basis for their own adaptions. What a rating system is, is a mathematical formula, with the philosophical purpose of attaching a number that represents a person’s skill, to the person.
This means that everyone has a rating that represents a person’s hypothetical “true skill”, even if the person has never played the game.
So how it works in regards to this, is that a rating system is designed to bring a player to that corresponding “true skill” rating, as closely as mathematically possible.
The Elo system (not ELO, it’s not short for anything, it was just named after its creator), did a good job of it. This is what WoW used as a model for its original arena rating system back in burning crusade, when all teams started on 1500 and what was lost by the losing team was what was gained by the winning team. Then people had their personal rating inside the teams, which eventually would catch up to the team’s rating if the person wasn’t a part of every game from the start of the team, but this wasn’t at the cost of the losing team.
However, there are other rating systems. There’s a system called Glicko RD (Glicko is again just named after its creator Mark Glickman, and RD stands for Rating Deviation) which introduced rating deviation to the mix. It’s similar to the Elo system, but in essence it’s designed to make a player reach the corresponding “true skill” rating faster than with the Elo system.
You can check out Mark’s website at http://www.glicko.net/
Functions
Right, so many of you seem to be under the illusion that more people = faster queues, which the speed of the queue isn’t necessarily contingent on.
To explain this, I first have to mention the basics.
In order for a matchmaking system in a game to match people the way it does, it literally only needs enough people to fill one side, and to fill the other side, and then a server configuration prepared for hosting the match, regardless if it’s hosted by players themselves, 3rd parties, or the game company itself.
These are the bare minimum required for it to work.
So when introducing things like rating, rating deviation (the MMR in WoW follows a rating deviation pattern), and other metrics, that’s when time starts being the cost.
Metrics
Right, so Blizzard is using the Glicko RD system (or Glicko Boost, I’m not sure, but it’s basically Mark’s revised version of the RD system he created in the mid-90s) as its basis now, that’s fairly obvious.
But they’ve also added so many other metrics, and seemingly constantly adjust it or have added an AI to handle it for them (company secrets, so only they can know).
Metrics like starting out in a rating range and then incrementally increasing the search parameters the more time you spend in the matchmaking queue.
Metrics like average wait times (which doesn’t make it go faster, when without it would’ve matched as soon as a match is found, it instead makes it average out for everyone so there are no outliers).
Metrics like win streaks, it can actually be controlled by admins to take things like win streaks into consideration vs. specific comps and also possibly match your comp vs. what you struggle against, which they can change arbitrarily at will.
They can also prepare different presets of configurations for different player pool sizes, and they can do a lot of things behind the scenes to all kinds of metrics which affects queue time.
Something they also do, is control the total amount of the rating pool. Before, there was each new team injecting more rating into the pool, but now when everyone starts on 0 it doesn’t really work like that. With this system it doesn’t even award the same points for winning as the opponents lost all the time, so when a person wins more than the opponents lost, then more rating is created and added into the total rating pool. When a person lose more rating than the opponents win, that’s when the system deletes rating from the total rating pool size.
So by manipulating the metrics in these ways and subsequently the matchmaking, they can keep the top rating levels steadier without a fluctuation that corresponds to the total amount of characters having injected rating into the pool size.
Which is how you got that crazy rating inflation in BfA s1, yet lower rating with more characters having participated in MoP and WoD for example.
There is a lot of movement of small numbers and metrics which can heavily influence the speed of the matchmaking, and the rating pool size. Which they’re constantly manipulating.
A matchmaking system can also be designed to fit a certain expected player pool size. Let’s say you have 20 different rating ranges in a system like this. The activity in each rating range would depend on the metrics and the size of the player pool. For example, sorting it into 20 rating ranges with 40 people wouldn’t really go anywhere, while if you would use a single rating range for the matchmaking between those 40 people then it would go faster. While 20 rating ranges between 100,000 might be too few ranges, and not accurately represent the skill differences. So you would get a wider skill differentiation in all of the ranges like this, than you would if there would be 200 rating ranges in the system.
Conclusions
So matchmaking systems and rating systems can be adjusted for optimal efficiency in different player pool sizes. Because as mentioned, the bare minimum it needs is just enough players to fill your side, and enough players to fill the opposing side, and room on a server to host it.
So the real effect of having more players, is that the gaps in skill differences in matchups are smaller on average. It does NOT however make it faster, per se. How fast it goes depends entirely on the design of the matchmaking system and what it’s optimized for.
https://www.microsoft.com/en-us/research/wp-content/uploads/2006/10/Game-Developer-Feature-Article-Graepel-Herbrich.pdf here’s an article which many of you will find interesting, if you care about things like the design of the rating system, and why player feedback is important.