1911 replies to this topic

[Skadi]

Have they finally locked you in the room with the server so you can kick it every time theres a desync?

[J0anna]

I know PGI is loath to give out specific details on matchmaking and elo. I was wondering is I could get a answer to a very specific question, when exactly does the matchmaker start to relax it's boundaries?

For example if I want to experience the "best" balanced matches, should I cancel my search at 15 seconds,30 seconds or longer? I'm willing to wait a long time for a game since this is a experiment, and I am well aware I may not get a match for a long while, but Russ has asked us to consider relaxing our choice of game modes to allow for better matchmaking, so I would like to see what the matchmaker thinks is better matchmaking. If you can't give an exact time, is canceling every 30 seconds fast enough to get what I'm trying?

Edited by Moenrg, 24 September 2014 - 12:40 PM.

[Shlkt]

Moenrg, on 24 September 2014 - 11:00 AM, said:

For example if I want to experience the "best" balanced matches, should I cancel my search at 15 seconds,30 seconds or longer?

Canceling your search early will not improve your results. The timed release valves do not apply to individual pilots and how long they've been sitting in the queue; rather, they apply to specific games that the matchmaker is trying to form. You're just as likely to be dumped into a release-valve match at the beginning of your search. The match may have been sitting in the queue for two minutes already before you clicked "Play Now", so there's very little correlation between your personal queue time and the quality of the matchmaking results.

[Lordred]

Dear Karl. There is a measurable difference in performance between having the game HUD on, and off when the fighting gets intensive.

Are there any plans to address this?

Thanks.
~Red

[990Dreams]

Lordred, on 25 September 2014 - 08:40 AM, said:

Dear Karl. There is a measurable difference in performance between having the game HUD on, and off when the fighting gets intensive.

Are there any plans to address this?

Thanks.
~Red

I second this. When in combat my FPS drops by at least 30% (at most I totally and forever de-sync from the match).

[p4r4g0n]

This was addressed by Matthew Craig in this post re: scaleform. From his comments, it does not look like it will be addressed any time soon or at least not till CW Phase 2 is out.

[Karl Berg]

Modo44, on 16 September 2014 - 03:47 AM, said:

Unfortunately, hit registration has been getting worse for a while now. I think it started with the previous server move, not the recent one. Things used to be rock solid for me last year -- even SRMs mostly worked when they still had the major issue. These days, a stable, fast connection and the client running smooth do not seem to be enough. (I have a ping of just under 130 with typically ~1 ping jitter and 100Mbps both ways, options set up to at least keep 30fps at all times.) Things got even worse with the recent desyncing problem -- now essentially any aimed weapon has issues. This is not just a high ping/inconsistent connection problem anymore.

On the other hand, matchmaking seems to be improving -- very slowly -- at least in the solo queue. More and more matches are not stomps, and I see pretty close ones way outside of US prime time. I guess relaxing the tonnage limits is helping?

Hrm, I see. With everyone focusing on the current December milestones, very little attention is focused on hit detection issues. I'll see what I can do to raise this issue, and keep it raised as a priority for investigation; but I think there is little chance of any serious time being allocated towards hit detection until after some of the major community warfare tasks are completed, and resources are freed up to roll off.

Good to hear matchmaker is improving! Some of the variance reduction changes were launched a while back, without me being entirely ready for it. I spent some time watching a few of the MWOPro streamers make their way through the matchmaker queue to see what the resulting impact was, and Elo standard deviation looked to be roughly half what it was before the changes.

Valore, on 16 September 2014 - 06:05 AM, said:

The main culprit seems to be lasers, with ballistic, PPC and SRM hits only affected to a lesser extent.

That does help narrow things down for us, when we get some time to look into this. Thanks Valore!

[Karl Berg]

Kageru Ikazuchi, on 16 September 2014 - 06:25 AM, said:

Now, please understand, this is based on my experience (~200 ping), but ever since server-authenticated hit effects went live, it has taken me about a quarter to a half a second to see the reticule go red even when I score a solid hit point blank.

While it may be "working intended" it is also very frustrating, and potentially very dangerous when you're waiting for the laser burn to stop before twisting.

Hi Kageru; that red reticule has actually always been delayed by ping, even prior to the server-side hit effects. The server side hit effects change also made explosions delayed until after the server verified the hit. This was done to improve players abilities to determine whether or not the server thought they had hit and dealt damage.

[Karl Berg]

EoRaptor, on 16 September 2014 - 10:35 AM, said:

I have a question about the matchmaker. If it's been answered before, please say so, and I will hunt further through this thread.

The group selection screen is currently set up to allow the following:

1-10 players have a restriction of 3/3/3/3.

This leads to matches where I see lances of 3 direwolves and a timberwolf, or 3 timberwolves and a direwolf, often on the same team.

Would not a more layered selection screen provide better matching?

1-4 players get 1/1/1/1
5-7 players get 2/2/2/2
8-10 players get 3/3/3/3

Hi EoRaptor,

Yes indeed, that would definitely improve the matchmakers ability to combine multiple groups to produce matches. So far though, we've tried to integrate threes and larger groups into the matchmaker with as minimal impact to player behaviours as we could manage. If we were to restrict weight class counts based on group size, that would pose a significant increase in player restrictions when grouping.

While it would almost certainly improve the quality of matches being produced, I would need design to ultimately make this call. There is a very definite trade-off here, and it would need to be considered very carefully; including soliciting player feedback to determine whether or not such a change would be well received in general.

[Karl Berg]

Bad Karma 308, on 17 September 2014 - 11:23 AM, said:

Karl,

1.) Several patches back the SLI option made an appearance but was DOA, and removed a couple of hours later in a hot fix. Is there any update from your guys on when it might return?

2.) Also, Lately I've noticed some strange occurrences with weapons on weird "Ghost" trajectories.

Often it is with PPCs. What I'll see is that when someone else fires a PPC, A PPC bolt will occur, but it will originate from where there isn't any mechs. Most times it will originate from a very high altitude from where there is no terrain or mechs even possible. And no it isn't from a Jump jetting mech. The bolt will often originate and terminate just like the bolt from the mech who fires it. But again its origin and termination are way off axis and do not correlate to the actual shot in anything but maybe its trajectory.

If this isn't something you guys know of let me know and I'll try to get some screen captures/videos of it happening.

Very Respectively

Hi Bad Karma,

I believe the current holdup for SLI is a driver level fix. I don't have an ETA for release.

On two, no, I've never seen or heard of that before! If you manage to capture some video of this for us, that would be extremely helpful!

Thanks!

[Karl Berg]

Jonathan Paine, on 17 September 2014 - 07:25 PM, said:

Dear Karl Berg.

First of all, I thoroughly enjoy your participation on the forums and your insightful comments. Furthermore, I loved what you did to the match maker. The single player queue tends to produce close and interesting games. The group queue was amazing for roughly one week, and has been going down hills ever since. Is there any chance you will look over the data and the code and try to figure out why we are back to rolls in one direction or the other as the modal type of game? And yes, I do have some screenshots to back this up.

Hi Jonathan,

Thanks for the compliments sir!

As for group queue, I'm sorry to hear that it's been degrading in match quality for you. The group queue is very much influenced by player behaviour, simply due to it's highly restrictive nature. For any group that drops into queue, there is just a tiny fraction of other groups in queue that could be valid match candidates, due to all the various constraints that must be met for a match.

I am very interested in seeing group queue improve in quality as well. The game mode vote will definitely provide some amount of improvement. I ran these simulations quite some time ago, but I don't have the numbers on hand to share with you right now. More aggressive changes might be able to do even better; but almost all of these changes will require some level of tradeoff for users. I will have to discuss any of these options in detail with design before we can announce or commit to anything.

[Karl Berg]

Shlkt, on 25 September 2014 - 07:16 AM, said:

Canceling your search early will not improve your results. The timed release valves do not apply to individual pilots and how long they've been sitting in the queue; rather, they apply to specific games that the matchmaker is trying to form. You're just as likely to be dumped into a release-valve match at the beginning of your search. The match may have been sitting in the queue for two minutes already before you clicked "Play Now", so there's very little correlation between your personal queue time and the quality of the matchmaking results.

This is correct. The release valves are a function of the oldest request currently assigned to any pending game. The matchmaker is constantly breaking and reforming potential matches around those oldests requests to determine if it can produce an optimal game given it's current resource set and constraints.

That said, if you yourself happen to be the oldest request, Elo doesn't really start loosening until after roughly two minutes of searching. By default, you enter the queue with around a 50 Elo range. That widens to an absolute maximum of around 1400 after 5 minutes searching for solo queue, although I don't think I've ever seen a match much older than about 3.5 minutes in solo queue.

p4r4g0n, on 25 September 2014 - 09:11 AM, said:

This was addressed by Matthew Craig in this post re: scaleform. From his comments, it does not look like it will be addressed any time soon or at least not till CW Phase 2 is out.

Yes unfortunately. Scaleform 3 is not the most optimized library in terms of render performance. Moving to scaleform 4 would provide significant improvements in terms of batched draw calls and a newer, better actionscript interpreter. Unfortunately porting CryEngine's UI subsystems to scaleform 4 would be non-trivial, and we don't have access to the source level modifications Crytek made to the scaleform libs when they did their integration.

[Cimarb]

Karl Berg, on 27 September 2014 - 04:06 PM, said:

Good to hear matchmaker is improving! Some of the variance reduction changes were launched a while back, without me being entirely ready for it. I spent some time watching a few of the MWOPro streamers make their way through the matchmaker queue to see what the resulting impact was, and Elo standard deviation looked to be roughly half what it was before the changes.

Thank you so much for focusing on deviation, and not averages between teams. It will not matter in CW, but it makes a huge difference currently!

[Jman5]

Hey Karl, I keep reading that the group queue matchmaker is struggling to juggle the varying team sizes, mechs, Elo, and game modes. Would it help the matchmaker if it could slot a single solo player into the group queue when you get an odd-man out scenario? So if Matchmaker has this 12 man on one side and an 8man and a 3man in queue, it could pull a solo player in that perfectly matches the mech/Elo needed. You could set a hard cap to no more than 1 solo player per team.

I think there are a lot of people out there who wouldn't mind occasionally being put into the group queue if it was a voluntary opt-in that was off by default. It would also allow players who might not be a part of any unit showcase their skills in the group queue. And as a nice bonus it would open the door for 11-man teams.

Thoughts?

Edited by Jman5, 27 September 2014 - 06:06 PM.

[Wintersdark]

Jman5, on 27 September 2014 - 06:06 PM, said:

Hey Karl, I keep reading that the group queue matchmaker is struggling to juggle the varying team sizes, mechs, Elo, and game modes. Would it help the matchmaker if it could slot a single solo player into the group queue when you get an odd-man out scenario? So if Matchmaker has this 12 man on one side and an 8man and a 3man in queue, it could pull a solo player in that perfectly matches the mech/Elo needed. You could set a hard cap to no more than 1 solo player per team.

I think there are a lot of people out there who wouldn't mind occasionally being put into the group queue if it was a voluntary opt-in that was off by default. It would also allow players who might not be a part of any unit showcase their skills in the group queue. And as a nice bonus it would open the door for 11-man teams.

Thoughts?

I'm definitely one. I'd absolute opt-in to the group queue. I much prefer the more organized nature of those matches. I *liked* having premades in the queue Back In The Day.

Don't force it, or enable it by default; but allow players to opt in.

[Kageru Ikazuchi]

Karl Berg, on 27 September 2014 - 04:11 PM, said:

Hi Kageru; that red reticule has actually always been delayed by ping, even prior to the server-side hit effects. The server side hit effects change also made explosions delayed until after the server verified the hit. This was done to improve players abilities to determine whether or not the server thought they had hit and dealt damage.

Thanks for the reply ... that is what I figured, but I am still curious ... approximately how quickly (in terms of ms + X + Y, where X is my ping, and Y is my target's ping) should I get the hit detection feedback (red reticule or hit/miss effects)?, and at what point does HSR throw in the towel (I seem to remember hearing 250 ping being the max rewind at one point or another).

Thanks for the dialog!

[slide]

I am pretty sure max rewind on HSR is 450ms.

[p4r4g0n]

Its been confirmed a couple of times in the past that HSR does not work for pings above 450ms as slide mentioned. It also doesn't work well, IIRC, if your connection is not stable i.e. bursts of packet loss or latency spikes.

[ShinVector]

p4r4g0n, on 28 September 2014 - 05:50 AM, said:

Its been confirmed a couple of times in the past that HSR does not work for pings above 450ms as slide mentioned. It also doesn't work well, IIRC, if your connection is not stable i.e. bursts of packet loss or latency spikes.

Nor does Lasers HSR work well at 232-235ms....
Not much jitter since the more to the OVH... but lasers hit detection is pretty horrid now...
Only way to hit anything reliably is:
1. You stand still.
2. Target stands still and hope he doesn't move at all for 1 second for the lasers to finish burning.

Ay, Caramba !



]

Edited by ShinVector, 28 September 2014 - 07:56 AM.

[shellashock]

Hope you will be able to make the switch to Scaleform 4 someday. Here's hoping Crytek will let you know what to modify if the day comes that you have time to implement it.