Slow down on mid-sized scenarios

[thewood1]

Here the Eagle vs Bear Scenario. Is a 250ish unit scenario. I get periodic slow downs where pules are consistently in 1500 - 2000 range. It seems to run between 18 and 50ms with spikes up to 150 ms for 5-6 minutes at a time. But then shoots to above 1000 ms for a minute or so. I am not sure what is happening in game to cause this.

State of the game

Build 551
All range circles off except air, showing range for selected unit only and no merging of range circles.
No-pulse mode on and hi fi on. I have tried both 1:1 and 1:5 time ratios. I have also tried hi fi off and that helped only a little.

My laptop is an A10 2.5MHz to 3.5MHz quad core (really only two good ones), 16Mb RAM, a good graphics card Radeon 85XX.

I have kept an eye on RAM usage and CPU temp. Nothing unusual.

I have attached a save just as it spiked to 2200 ms pulse.

Just a note that 1000 units scenarios have previously run fine. I only noticed this on 546 but it might have been happening earlier.

[mikmykWS]

Looks like the navigator (the navigator code is calc intense while plotting). Dull the sharp corners and you don't see such intense spikes. We'll take a look.

Mike

[NakedWeasel]

I'm honored that you are using my scenario as your test basis! I also saw dramatic slowdowns with B551. I'm suspecting the unit pathing is the problem. I've included my latest build of the scenario, tell me how it performs with 551.

[thewood1]

It's a cool program. But I suspect it also has something also to do with jamming/being jammed...basically a lot of ECM going on.

[thewood1]

So I have used Battle of Latakia since v1.00 as a quick benchmark. It has typically run a 2-4ms pulse. And that was on an older and slower laptop. Last time I ran it on this one at 536, I remember it never going above 2ms. No I run it on 551 and it runs 4-50ms.

There are less than 20 units here with very straight forward paths.

[NakedWeasel]

Here it is being very naughty. B551.

[thewood1]

So I went back to 1.00 and ran some tests on Operation Bass Drum vs 551.

Average pulse time at 5:1 time ratio for the first hour of game time: 1.00 - 65ms...1.04 RC3 Build 551 - 280ms.

All the same settings and 1.00 would spike to 100ms and 551 would spike to 450ms.

I used pulse time for both to set a level playing field. When I turned non-pulse on in 551, pulse went to about 200 average, but the time ratio was only a little better than 2:1, even though it was set at 5:1.

edit...

Just ran Operation Bass Drum in 1.03. Basically ran the same in pulse mode. It ran much faster in non-pulsed.

Ran it in 104 RC 536 and it is a tad bit slower than 1.03, but not much. As soon as I went back to 551 though, it was struggling.

[mikmykWS]

Very useful thank you.

Mike

[thewood1]

I am here to serve.

[Dimitris]

Just to clarify something.

Since B536 we had to add a lot of extra checks and hardening measures in order to remedy a growing series of inconsistent, non-reproducible crash bugs that were ruining the experience of a significant number of players. This means that a certain performance drop is inevitable. Think of it as up-armoring a tank. It loses some of its pep, but it gets you through downtown Baghdad in one piece.

The question now is (a) whether the performance drop is solely due to these extra stability measures or there are other issues involved, and (b) what can we do to restore as much of the lost performance as possible while also retaining the vastly improved stability that we (and hopefully you too) have observed on the last 1-2 public builds. We are looking into this intensely in order to find answers to these questions.

Thanks for your feedback and (hopefully) your understanding.

[thewood1]

No prob, but I got to say, I will most likely continue to play the 536 build. If I am getting that performance on only a couple hundred units, the game is darned close to unplayable. Its not just effecting the pulse, the entire interface is slowing down. Including map zoom, map scroll, db viewer, menu selection, etc.

If it were just the pulse counts, I might be able to live with it as a balance to new features. But the entire game is running like mud. I know you'll do what you can, but I have limited gaming time and I have to decide how I want to spend it, based on how unfrustrating a game is.

[Dimitris]

ORIGINAL: thewood1

Here the Eagle vs Bear Scenario. Is a 250ish unit scenario. I get periodic slow downs where pules are consistently in 1500 - 2000 range. It seems to run between 18 and 50ms with spikes up to 150 ms for 5-6 minutes at a time. But then shoots to above 1000 ms for a minute or so. I am not sure what is happening in game to cause this.

State of the game

Build 551
All range circles off except air, showing range for selected unit only and no merging of range circles.
No-pulse mode on and hi fi on. I have tried both 1:1 and 1:5 time ratios. I have also tried hi fi off and that helped only a little.

My laptop is an A10 2.5MHz to 3.5MHz quad core (really only two good ones), 16Mb RAM, a good graphics card Radeon 85XX.

I have kept an eye on RAM usage and CPU temp. Nothing unusual.

I have attached a save just as it spiked to 2200 ms pulse.

Just a note that 1000 units scenarios have previously run fine. I only noticed this on 546 but it might have been happening earlier.

Did you try this under B536? Seems to be pretty slow under that too.

[mikmykWS]

Bummer. You're missing out.[:)]

We'll continue to investigate the speed issue. Thanks for your time and effort.

Mike

[chesmart]

Sunburn on my system build 536 uses 50% of performance whilst 551 uses only 10%

[mikmykWS]

Ok just ran Duelists. Largest spike was 100+ during the Shipwreck raid. Not even close to unplayable.

Mike

[mikmykWS]

Guys can you let us know how things go with your games this weekend? We want a good sample population on this.

If you can run the scenario above but then run others as well.

Thanks!

Mike

[ryszardsh]

Mike

one thing is clear to me at this point: 551 has more game speed variations. I run the game on an Intel NUC with 8gb of RAM, built in video and w8.1 Prior to 551 I saw slow downs but now I see faster running generally (pulse times are clearly reduced) but occasional spikes of pulse time >2000. Noticed in the carrier/bear above, in a med scenario I am playing with and in the Chingjin strike scenario too. Will run a few more different size scenarios over weekend in general so far, game runs faster in 551 but when slowdown hits, amplitude of slowdown is maybe 50% greater than last RC but frequency of slowdown (as measured by pulse time) is significantly lower, maybe 50% or more (I can only estimate the frequency) As a result the game feels choppier - the performance dips are much more noticeable b/c of the lower frequency and the somewhat faster overall speed.

RAS

[mikmykWS]

Have had 2 internal beta testers check in and 2 or 3 guys in Baloogan's chat room check in on this. There are no claims of it being unplayable or deathly slow. Will continue to test and monitor.

Mike

[Feltan]

Ran some tests on the Duelist Scenario. No problems at all.

Here is the baseline measurement of my system with no apps running.

Regards,
Feltan

[Feltan]

Here is the measurements with CMANO started, but no action.

Regards,
Feltan