[REPORTED]Planes piling up at airbases

[FlightControl]

APC-001 - Sukhumi-Babushara

 

Overall: Not good

Time: 01:49

 

-- 9 planes hold for about 20 minutes before one lands.

-- The last 2 planes held for 20 minutes before landing!

-- Destroyed plane on runway! 9 planes removed.

-- Test failed.

 

APC-001 - Takeoff and Landing - Sukhumi-Babushara.miz

[FlightControl]

APC-001 - Senaki-Kolkhi

 

Overall: Very good!

Time: 01:01

APC-001 - Takeoff and Landing - Senaki-Kolkhi.miz

[FlightControl]

APC-001 - Kutaisi

 

Overall: Very good

Time: 01:00

APC-001 - Takeoff and Landing - Kutaisi.miz

[FlightControl]

APC-001 - Kobuleti

 

Overall: Very good

Time: 00:53

 

++ This was one of the most efficient airbases.

APC-001 - Takeoff and Landing - Kobuleti.miz

[FlightControl]

APC-001 - Batumi

 

Overall: Good

Time: 01:43

 

-- The last 3 airplanes were holding more than 50 minutes before landing.

++ The best performing airbasein terms of CPU, despite the 3 landing holds.

 

APC-001 - Takeoff and Landing - Batumi.miz

[FlightControl]

APC-001 - Soganlug

 

Overall: Very good

Time: 00:42 (fast!)

 

- Some planes were holding a bit too long.

- Landed airplanes at the airbase were removed due to limited parking spots. That's okay!

 

APC-001 - Takeoff and Landing - Soganlug.miz

[FlightControl]

APC-001 - Tbilisi-Lochini

 

Overall: Very good

Time: 00:53

APC-001 - Takeoff and Landing - Tbilisi-Lochini.miz

[FlightControl]

APC-001 - Vaziani

 

Overall: Very good

Time: 00:58

APC-001 - Takeoff and Landing - Vaziani.miz

[FlightControl]

In summary of APC-001 Test scenarios, some of the findings:

1. Planes crash into each other while taxiing.

2. Planes stop taxi and never depart.

3. Planes hold sometimes for a very long time without apparent reasons.

4. Planes are removed after landing. Just vanish.

5. Some airport operations consume high CPU.

6. Efficiency of airbases differ a lot in Caucasus.

7. Beslan is really bad.

8. Anapa is really bad.

 

9. There is also good news, some airbases are really efficient in the optimal scenario (no destroyes etc).

 

Some other remarks ...

 

- Why do landed planes after parking remain on the parking spot for hours? Although there isn't a performance impact (it seems) it is a problem for airbase operations.

 

- I noticed in the last DCS 1.5.7 release, although airplanes were destroyed, the landing and taxiing just continued until a block happened. Has this behaviour changed over the last release patch? Because I don't see an update note of that. In previous releases any destroyed airplane will halt airbase operations immediately.

 

The next test scenario will be about testing damaged planes returning to airbases... Landing behaviour will be observed.

[FlightControl]

One may ask; why all this and why such a fuss?

 

Well:

 

1. Airbase operations are crucial to effective air defenses.

2. Intercept must be fast, so airplanes takeoff should be very efficient.

3. Damaged airplanes should be a priority, but only if no urgent takeoff.

4. Returning airplanes should have the lowest priority, only in urgent cases not, like out of fuel, or severely damaged.

 

I know that these are very demanding requirements.

 

I know that building a solution for these is not an easy feat, to be honest, the current APC is already good!

 

But unfortunately not good enough. If possible, could ED have a look at these issues? I would really like to have a look at the APC logic to see where optimizations are possible. Not sure what are the principles defined and logic built that is in place at the moment.

 

- Deadlocks, timings, and prioritization are key parameters in such a process. I know this is not easy and very difficult to get right. Many pitfalls and exceptions can occur. A good APC development may take quite some time, as the audience is big and the exceptions can be huge.

 

But still... Are there any improvements possible on what the test results show?

[BIGNEWY]

Thanks for the report, I will mention it to the ED team on Monday.

 

A small point, if you don't use MOOSE to spawn the aircraft it all seems to work fine, I am no script expert so will leave it to those that know to take a look.

[=Mac=]

I think it can go without saying: AI in DCS isn't all that Intelligent. As a consequence, the developers have taken short cuts to get some semblance of useful AI (such as bandits seeing through trees and mountains.)

 

Years ago, I was an AI researcher and was often surprised at how a beautifully functioning AI system can suddenly, and strangely, do some of the silliest things for no apparent reason yet still follow coded logic either given or acquired on its own. Human beings have had 200,000 years to develop biological intelligence enough to allow high quality email texting. Yet, even after all these years, traffic at most metropolitan centers are hilarious studies in Gordian Knots.

 

The news for AI isn't all that bad, though. From Boston Dynamics, we have learned how to get physical machines to walk and run. We have also discovered what is now called Deep Learning and with this concept in mind, researchers have managed to explore several different artificial worlds such as Chess and the Chinese game of Go.

 

DCS has a long way to go before an artificial ATC can suffice in place of a human one. But it still is the absolute top in the field of flight simulators.

 

Nevertheless, I urge all developers to "keep on keeping on!" And I thank you all for doing what you have thus far and what you plan to do in the future.

[FlightControl]

Thanks for the report, I will mention it to the ED team on Monday.

 

A small point, if you don't use MOOSE to spawn the aircraft it all seems to work fine, I am no script expert so will leave it to those that know to take a look.

 

MOOSE is only used for the spawning, nothing else.

[BIGNEWY]

MOOSE is only used for the spawning, nothing else.

 

Yes that is my point, If I place 10 aircraft at Anapa the normal way the airfield handles them correctly

[FlightControl]

Yes that is my point, If I place 10 aircraft at Anapa the normal way the airfield handles them correctly

 

Maybe, and when planes are dynamically spawned the operations fail sometimes.

Most servers spawn dynamically, as they run GCICAP functionality or other spawning logic.

I've been soft on the tests. Now planes are spawned every 3 minutes. If we try every 2 minutes and more planes, some airbases come to have real issues.

 

But that is for the next pile of tests to show.

The tests below are the "optimal scenario" tests.

Next come out of fuel, damages, airbase crashes...

 

I will also do a scenario without spawning but just placing 10 planes at an airbase and complete the findings.

 

We are suffering these things for years and to my knowledge never a good report was made on these things, very sporadic issues were raised though. The diagnosis is the first, then comes the cure... If we dont put some focus on this, nothing will change, right?

ED will assume all is okay, while maybe it isn't.

 

Just note that I KNOW this stuff is not easy to optimize.

So all respect to the team if they are okay to consider these matters for an investigation.

 

 

FC

[BIGNEWY]

No worries we will take a look, but also understand there is a long list of other stuff with a much higher priority.

 

I have PM'd a ED team member to highlight it already, so will get some feedback hopefully

 

but no promises :)

[Major Puncake]

I have had issue like that with bombers it is quite funny to watch them crash into each other as they try to find a parking place

[=Mac=]

I have had issue like that with bombers it is quite funny to watch them crash into each other as they try to find a parking place

 

This crash activity is completely normal for the Toronto area's highways and grocery store parking lots. Please make sure there are no Toronto politicians on your troubleshooting board!

[Pikey]

Is it me or is the Russian driving worse that the Georgian driving? j/k :)

 

There's defintiely a link to worse performance on the Russian airbases though, I knew some of them, but not this level of detail. Like one of the Krasnodar's has a circular path where taxiing aircraft end up facing each other under certian circumstances. Those should be sorted, you don't need any scripts to experience it, but scripting helps demonstrate the issues faster than any other way.

[BIGNEWY]

Is it me or is the Russian driving worse that the Georgian driving? j/k :)

 

There's defintiely a link to worse performance on the Russian airbases though, I knew some of them, but not this level of detail. Like one of the Krasnodar's has a circular path where taxiing aircraft end up facing each other under certian circumstances. Those should be sorted, you don't need any scripts to experience it, but scripting helps demonstrate the issues faster than any other way.

 

A lot of the old taxi problems are being looked at as part of the new terrain testing

[Grimes]

Jeez spent the day away from the forums and the thread exploded in size.

 

I guess I'll add comments on it since I asked you to post a mission. Didn't need to do it for each base, but still appreciated. Some of these issues have been present in one form or another going back a long time. With the scripting engine and just being more prone to spawning more aircraft at multiple times after mission start can simply exasperate certain issues. Simply put certain bases suck at handling take-off and landing operations at the same time. Most are pretty good for take-off, but overall landing behavior is sub-optimal. Depending on base design problems will arise with AI taxiing back to parking and others taxiing for take-off.

 

 

We are suffering these things for years and to my knowledge never a good report was made on these things, very sporadic issues were raised though. The diagnosis is the first, then comes the cure... If we dont put some focus on this, nothing will change, right?

ED will assume all is okay, while maybe it isn't.

 

Plenty has been reported on it over the years and especially as of late. A major part of terrain testing is checking AI behavior at airbases, which with new maps can be quite extensive. Most of the time its simply to manage issues with a specific base to fit within current behaviors. A great example was Beslan where a flight of 4 aircraft used to taxi down the runway to the end to turn around only for lead to get stuck or crash into -3 because there simply isn't space. Thing is Beslan was the first base of that design in DCS and the AI logic for taxi behavior didn't really know what to do with it. IMO it needs to be overhauled more than anything, but I don't get to make those decisions. The two biggest issues on that front in my opinion are the lack of system aware of all base operations and blocking system to rigidly dictate if an AI can or cannot transit on a given block of taxiway. With those two issues you can easily break any base that has a taxiway used to travel in either direction simply with proper timing and two aircraft.

 

Re your results.

 

Beslan, Sochi, Novo, and Ghelen didn't surprise me as those have been the more problematic of bases in game. The U-turn bug will hopefully be fixed on the updated map and is more or less present because ED stopped updating the old map.

 

Anapa, Krymsk, Mineral, Gudauta, Sukhumi, Kras-P, Kras-C, and Maykop results surprised me and I'll check those out more in depth.

[FlightControl]

@Grimes, fantastic. Thank you for taking your time going through the report(s) and interpreting the results. I know plenty had been reported in the past.

 

This is only the first batch of tests. More tests to come.

 

I hope the table in post provides a good overview.

 

https://forums.eagle.ru/showpost.php?p=3236467&postcount=12

 

For years we are suffering these things. A lot of these issues were reported but pulled to the back of the fix queue or it never got attention by the team. ED may or may not take this as a (positive) signal. Trying to collect the facts through objective measurements and document by demonstration.

Test missions are made, which took me about 6 hours to test, observe, document findings and report. I hope ED will spend some time in this in the near future.

 

FC

 

And please to the community, please let us try to focus instead of deviating to side discussions.

Edited September 5, 2017 by FlightControl

[funkyfranky]

Yes, I can confirm some of this happening also on NTTR. When several aircraft arrive at the same airport simultaneously for landing, they get into each others way and neither manages to land.

 

More specifically, one aircraft seems to get landing clearance but when on final, other aircraft which are holding get into its way. The aircraft which wanted to land breaks off its attempt. This can go on for a long time until eventually one aircraft successfully manages to land.

 

@FlightControl: Thanks for all your tests! This is valuable information.

[FlightControl]

@ED Is there any feedback what actions are planned to solve these malfunctions?

[BIGNEWY]

@ED Is there any feedback what actions are planned to solve these malfunctions?

 

As you know FC there is a big list, we will know only when ED put it in the patch notes.