No replies to this topic

[capnjosh]

Matchmaker and Server Browser changes are a hot topic.  Oh yes indeed :)

 

I've got some numbers and some tweaks I think we'll all agree are in order.  The goal right now for matchmaking is to tweak things enough that we can make out significant differences within a few days.  Then, we can adjust from there.  Of note, I have built out some consistent data sets we can look at over time to gauge what *actually* happened.

 

Some of these charts may not be screamingly clear enough.  Forgive me for now; I'll get better with time.

 

First up: We reduced the threshold that allows joining matches from the Server Browser.  We lowered the range from 500MMR difference to 300MMR difference.  More users experienced more cases where they couldn't join the selected match. This is also a slightly higher % of the population.  In graph form (the depressed section is actually due to data sampling reducing the raw number of data points, so ignore that part):

 

 

 

Next up: Differences between winning and losing teams (keep in mind these are indicators over time).  This indicates the matchmaker changes 2 days ago did have a tangible impact.  Winning teams on average have a higher kill spread within the team, and winning teams are getting more kills overall.  And, yes, I'd prefer to see that winner-loser kill gap tighten over time:

 

 

 

Next up: % of Matches with various spreads in kills.  In short, the changes have resulted in more matches having larger overall kill spreads between the best and worst players:

 

 

 

What do we do from here?

 

1. Expand the MMR threshold back to 500, which is what it was before.

  -Granted, relatively few players actually encounter this.  However, for those that do, it's *really* annoying.  We'll leave at this value through the weekend and then review the data that comes in.

2. Increase the time we give the matchmaker to expand its criteria in search of suitable matches.

  - This may seem a bit unexpected, but what has been happening with the 30 second expansion time is that each expansion iteration is pretty large, netting a bunch of servers all of a sudden.  In that case, the code is having a hard time telling *which* of those servers is the "best".  By effectively reducing the possible number of servers it can net in the first iteration it actually gets some valid servers to match a player to, we can be more certain of getting the closest possible server.

  - I've increased it to 90 seconds (this means you will get a match at most 90 seconds after entering the matchmaking queue)

3. Watch the data and then iterate.

  - If we don't get satisfactory results over the next few weeks, then we will have to identify more drastic action that will require client and game server builds.  I'm hoping we can at least see a tangible overall improvement in things before then though.

 

 

What happened over the last 7 days:

1. Verified solution to load-related slowdowns

  - we should not have any more big drops in CCUs, even under heave load

2. Pioneer Private Server roll-out was completed.  Support is ongoing.

  - yes, those mice are going out soon - I'm taking the first batch to the post office today

3. First changes to the Matchmaker were implemented

  - Tuning the existing system will take about 2 weeks - we need enough time after each iteration to gather enough conclusive data

4. We built our first functioning Hawken client.  To make it more of a challenge, we built a 64-bit client too.

  - We need to roll back some changes we're seeing in there, and it seems that may take some decent effort (for both client and server), but the point is that I was able to log in with a client built from source :)

  - The original estimate for an actual client build was roughly 4-6 months after we acquired Hawken, so we're still a good 2-3 months out.. just to set some expectations...  And, yes, it's ok to vent frustration that it's not faster/sooner/better/bigger/badder/full-of-win ;)

5. Boost Items are now at a full 100% boost (up from 50% boost)

  - a long time ago they were decreased to 50%... anyone have some juicy stories surrounding why?

 

 

What's happening this next week:

1. Get the mice all shipped

2. Start moving build servers to a permanent location

  - no, I do not like the idea of having servers sitting on desks

3. Matchmaker tweaks

4. A new mech package

5. Let's get some more items released - what do you think of that?