Board index FlightGear Development Aircraft

Announcing NEW Boeing 777 'Seattle' - Pre-Release

Questions and discussion about creating aircraft. Flight dynamics, 3d models, cockpits, systems, animation, textures.

Re: Announcing NEW Boeing 777 'Seattle' - Pre-Release

Postby Hooray » Sat Feb 01, 2014 2:14 pm

that's a "feature" - I am interested in bugs, especially PERFORMANCE
Please don't send support requests by PM, instead post your questions on the forum so that all users can contribute and benefit
Thanks & all the best,
Hooray
Help write next month's newsletter !
pui2canvas | MapStructure | Canvas Development | Programming resources
Hooray
 
Posts: 11329
Joined: Tue Mar 25, 2008 8:40 am

Re: Announcing NEW Boeing 777 'Seattle' - Pre-Release

Postby KL-666 » Sat Feb 01, 2014 9:30 pm

Just noticed something interesting about memory consumption:

Most planes have at startup:
- fgrun: 200-300 Mb
- fgfs: 600-700 Mb
- total: about 1 Gb

This 777 has at startup:
- fgrun: 1000 Mb
- fgfs: 1300 Mb
- total: about 2.3 Gb

My experience is that during flight more memory is needed for other things. So it is better to have less memory used initially. Especially the idle fgrun should not take so much memory away.

Kind regards, Vincent.
KL-666
 
Posts: 784
Joined: Sat Jan 19, 2013 1:32 pm

Re: Announcing NEW Boeing 777 'Seattle' - Pre-Release

Postby Hooray » Sat Feb 01, 2014 11:31 pm

that's interesting - how exactly did you measure memory consumption ?
You may also want to check out: http://wiki.flightgear.org/Subsystem-le ... FlightGear

Regarding fgrun, it makes zero sense to occupy all that memory when starting FG, things should be freed (deleted) prior to forking the FG process - feel free to file a feature request, it should be a straigthforward to clear memory in heavy instances, such as the 3D model preview- or just disable/hide the widget.

It would be also interesting if fgrun is leaking anything
Please don't send support requests by PM, instead post your questions on the forum so that all users can contribute and benefit
Thanks & all the best,
Hooray
Help write next month's newsletter !
pui2canvas | MapStructure | Canvas Development | Programming resources
Hooray
 
Posts: 11329
Joined: Tue Mar 25, 2008 8:40 am

Re: Announcing NEW Boeing 777 'Seattle' - Pre-Release

Postby KL-666 » Sun Feb 02, 2014 1:54 am

Hello Hooray,

It in not exactly measuring internal memory, but watching the process memory in the os of fgrun and fgfs.

In case of fgrun view plane x and see the process mem, next view plane y and see the process mem grow, and go back to plane x and see the process mem shrink to the original. Repeat this a couple of times to be sure.

Same for fgfs, start with plane x, do nothing and note the process memory. Next start with plane y in the same place and note more process memory in use. Then repaet to be sure.

Kind regards, Vincent
KL-666
 
Posts: 784
Joined: Sat Jan 19, 2013 1:32 pm

Re: Announcing NEW Boeing 777 'Seattle' - Pre-Release

Postby Hooray » Sun Feb 02, 2014 3:44 am

understood, can you post the htop/top or vmstat numbers for a few aircraft so that we can take a look ?



Regarding ND performance: I just checked it again: https://gitorious.org/fg/fgdata/source/ ... y.mfd#L176

The only real candidate that may be contributing to lag is actually the "route" layer, which hasn't yet been ported, and which isn't exactly straightforward to port - it's one of the more difficult layers, and not really suitable to be directly ported, so I am not sure if we will make it in time for 3.0 - we would basically need to split it into 2-3 different layers (waypoint, legs etc) and then update those selectively - currently, the heuristics are still "dumb" and still using removeAllChildren() which isn't exactly canvas-friendly.

Last time I looked at it, I wasn't entirely sure how to best port it without causing breakage, also because I am not exactly sure how to test it reliably - this would need to be coordinated with Gijs or Hyde, so that we can directly act if necessary.

So it would be up to Gijs and Hyde to document the existing code and clean it up a bit (or add sanity checks), so that we can take a look and try to generalize it in order to port to MapStructure - currently, it's still using the old method, i.e. algorithmically it's unoptimized and may cause stutter.

I am also not sure if MapStructure itself is sufficiently flexible to support this particular layer without any major additions ...

There are a few other layers still using the old removeAllChildren() logic, so these MAY contribute to lag, but not in any significant fashion - obviously, this is affected by the frequency of the running update loop - which defaults currently to 0.05 :?

Basically, we just need people to look at the layers specified in "navdisplay.mfd" - any layer that lacks the isMapStructure:1 field, is still an old, inefficient, layer and needs to be ported - analogous to the FIX/NDB/VOR or DME layers, which can serve as examples.

However, those other layers should be straightforward to port by aircraft developers like the 747/777 teams, just based on working through the MapStructure discussion, or even just looking at other ported layers.
Please don't send support requests by PM, instead post your questions on the forum so that all users can contribute and benefit
Thanks & all the best,
Hooray
Help write next month's newsletter !
pui2canvas | MapStructure | Canvas Development | Programming resources
Hooray
 
Posts: 11329
Joined: Tue Mar 25, 2008 8:40 am

Re: Announcing NEW Boeing 777 'Seattle' - Pre-Release

Postby Hooray » Sun Feb 02, 2014 5:43 pm

Philosopher wrote in Thu Jan 23, 2014 4:16 am:STA and WPT worked just fine, displaying instantly, while ARPT took a while to load the first time but worked fine subsequently.



we could add a callback to each controller that would have the only purpose of doing "cache-building-queries" while booting - i.e. to make sure that the navdb/sqlite cache is up and running once the sim is started - so no noticeable delay at runtime, but probably 1-2 seconds longer startup times - some russian aircraft developer is doing similar things to avoid runtime lags, i.e. by using a piece of Nasal code to iterate through all views - so that there's no longer a delay later on.
Please don't send support requests by PM, instead post your questions on the forum so that all users can contribute and benefit
Thanks & all the best,
Hooray
Help write next month's newsletter !
pui2canvas | MapStructure | Canvas Development | Programming resources
Hooray
 
Posts: 11329
Joined: Tue Mar 25, 2008 8:40 am

Re: Announcing NEW Boeing 777 'Seattle' - Pre-Release

Postby Philosopher » Sun Feb 02, 2014 5:54 pm

That kinda happens already, at least for other layers (VOR, DME, etc. IIRC). But yeah, I think Hyde's comment says that caching disappears after moving too far, so there's limited use to it. Just supporting incrmental searches would be the best solution I can see.
Thanks,
Philosopher
(inactive but lurking occasionally...)
Philosopher
 
Posts: 1590
Joined: Sun Aug 12, 2012 6:29 pm
Location: Stuck in my head...
Callsign: AFTI
Version: Git
OS: Mac OS X 10.7.5

Re: Announcing NEW Boeing 777 'Seattle' - Pre-Release

Postby Hooray » Sun Feb 02, 2014 6:00 pm

I think Hyde's comment says that caching disappears after moving too far, so there's limited use to it. Just supporting incrmental searches would be the best solution I can see.

See:
Subject: How to display Airport Chart?
zakalawe wrote:Regarding spatial queries, on the nav-cache side, delta queries would be complex to support. What the C++ NavDisplay does is keep a persistent list which is updated infrequently - only when a setting changes (range, view options config), or when the aircraft moves > 1nm.


We did talk about supporting min/max queries as a workaround back then, and I cannot see anything making this difficult or problematic - but I may be missing something.
Zakalawe mentioned in the tracker that he's using a "return-early" approach to "time out" - and that this could be easily added here, too.
We could also use split-frame loops to process each set of results directly per frame. It's still all single-threaded anyway ...
Please don't send support requests by PM, instead post your questions on the forum so that all users can contribute and benefit
Thanks & all the best,
Hooray
Help write next month's newsletter !
pui2canvas | MapStructure | Canvas Development | Programming resources
Hooray
 
Posts: 11329
Joined: Tue Mar 25, 2008 8:40 am

Re: Announcing NEW Boeing 777 'Seattle' - Pre-Release

Postby KL-666 » Mon Feb 03, 2014 2:37 am

Hooray wrote:

understood, can you post the htop/top or vmstat numbers for a few aircraft so that we can take a look ?


This is so easy to do, that you better do it yourself, so you can get any aircraft number you wish. Open fgrun and a process viewer, and just click on any aircraft and see the process mem of fgrun change.

In the mean time i did some tests, from which i conclude that the aircrafts do not particularly contribute to excessive mem use (although the less base mem they use, the better). There seem to be no leaks. Did not use flight director btw. But actually i am not really worried about it's mem use. The glass panels are canvas too, are they not?

I wrote down some remarks on fg3 mem policy in a new thread:

http://forum.flightgear.org/viewtopic.php?f=68&t=21984
KL-666
 
Posts: 784
Joined: Sat Jan 19, 2013 1:32 pm

Re: Announcing NEW Boeing 777 'Seattle' - Pre-Release

Postby Hooray » Mon Feb 03, 2014 2:53 am

I asked you to do it, because I cannot do it obviously, I am not using a computer currently that would allow me to run FG, that is why.
Besides, it does make sense to let different people provide these numbers, using different OS/platforms and architectures.
Please don't send support requests by PM, instead post your questions on the forum so that all users can contribute and benefit
Thanks & all the best,
Hooray
Help write next month's newsletter !
pui2canvas | MapStructure | Canvas Development | Programming resources
Hooray
 
Posts: 11329
Joined: Tue Mar 25, 2008 8:40 am

Re: Announcing NEW Boeing 777 'Seattle' - Pre-Release

Postby KL-666 » Mon Feb 03, 2014 9:44 pm

@hooray

Sorry, did not realize that you can not fly. Got some numbers taken on TLPL direct after startup. There is much water around the island, so not much scenery to interfere. The values are in Mb:

Aircraft, fgrun, fgfs

717, 208, 588
737ng, 157, 588
747-400, 235, 706
777, 1027, 1221
a320-family, 307, 651
a320-neo, 229, 671
a380, 262, 645

At the moment i am on fg3.0rc2 on win 8.1


@hyde

There is a very minor issue, but i wanted to mention it:

The roll knob of V/S rolls correct, but the mouse rolls opposite. Could you reverse the mouse roll for V/S?
KL-666
 
Posts: 784
Joined: Sat Jan 19, 2013 1:32 pm

Re: Announcing NEW Boeing 777 'Seattle' - Pre-Release

Postby zakalawe » Tue Feb 04, 2014 10:16 am

The new cockpit looks really nice, but for me the frame-rate is quite a bit less than before, especially if I try to look at the pedestal. Certainly not possible to make a landing at EHAM or KSFO, and keep FPS above 20, let alone 30.

I would suggest, to provide some way to use the previous cockpit model as an option? Eg make two different -set files, which load different cockpit models? Since I would like to have the new cockpit and fly, but right now I must choose either 'looking' or 'flying'!
zakalawe
 
Posts: 1152
Joined: Sat Jul 19, 2008 4:48 pm
Location: Edinburgh, Scotland
Callsign: G-ZKLW
Version: next
OS: Mac

Re: Announcing NEW Boeing 777 'Seattle' - Pre-Release

Postby Hooray » Tue Feb 04, 2014 4:12 pm

I guess it would be really useful to investigate eventually why certain cockpits/3D models are having such a heavy impact on some platforms, while others -more complex- apparently don't on different hardware. For starters, we could benchmark updating cockpit objects to determine which ones are taking lots of time to update (in comparison)
Please don't send support requests by PM, instead post your questions on the forum so that all users can contribute and benefit
Thanks & all the best,
Hooray
Help write next month's newsletter !
pui2canvas | MapStructure | Canvas Development | Programming resources
Hooray
 
Posts: 11329
Joined: Tue Mar 25, 2008 8:40 am

Re: Announcing NEW Boeing 777 'Seattle' - Pre-Release

Postby KL-666 » Fri Feb 07, 2014 9:49 pm

Hi Hyde,

Flying around a bit more with the 777, i really appreciate the fdm, working very natural in trimming for speed and throttle for alt.

But i also noticed that i do not get DME on VOR's. Where in other planes i get it at 150-200 mi on certain VOR's, i now barely get DME at 30 mi from any VOR. Can you look into that?

Kind regards, Vincent
KL-666
 
Posts: 784
Joined: Sat Jan 19, 2013 1:32 pm

Re: Announcing NEW Boeing 777 'Seattle' - Pre-Release

Postby Hooray » Fri Feb 07, 2014 9:57 pm

That's a bug in the ND framework that Gijs recently found, it's fixed already.

Image
Please don't send support requests by PM, instead post your questions on the forum so that all users can contribute and benefit
Thanks & all the best,
Hooray
Help write next month's newsletter !
pui2canvas | MapStructure | Canvas Development | Programming resources
Hooray
 
Posts: 11329
Joined: Tue Mar 25, 2008 8:40 am

PreviousNext

Return to Aircraft

Who is online

Users browsing this forum: amue and 34 guests