[11193][performance] Visibility updates and relocations

[SilverIce]

Hi,

This patch reduces amount of visibility update and AI relocation notification operations -- in this way you will get some performance speedup.

In original sources these functions called too frequently - this wastes power of CPUs for nothing.

We should filter redundant visibility update operations.

Some statistics: optimization decreases amount of visibility update operations in 8-10 times(expected decrease in CPU utilization about 15-20%).

Currently these are two different ways of how it should be optimized.

The first way:

Visibility and AI notification operations processed each N milliseconds. To avoid CPU spikes these operations processed simultaneously.

Possible problem: as a result of fixed time interval between visibility updates,

its impossible to increase frequency of these operations for fast moving players/creatures.

branch1: https://github.com/SilverIce/mangos/tree/visibility_optimization_old_way

Second way:

Can't say too much. No problems with visibility of fast moving units: frequency of visibility updates scales with unit's speed.

branch2: https://github.com/SilverIce/mangos/tree/visibility_op2_simplified

Note: added few new config setting to find optimal parameters, so don't forget update your config files

Also, these config options could be reloaded during runtime: you can experiment with these options without restarting

I think second way is more progressive, but I don't know performance difference.

Please, test the patches, feel free to comment, report test results..

[Ambal]

Guys, I suggest you to not miss this patch as it is essential to greatly improve mangos performance :rolleyes: Test branch2 first since we need to know how bigger the bonus it gives + it does not have issues with fast moving objects like branch1 does.

[kero99]

I go to test Branch2 at weekend with 2000+ players... i will post a feedback =)

[kero99]

Hi Silverice, i try to apply patch in my production server but code is soo much outdated and soo much merge/pickup conflict with last rev. i could try to resolve all conflict but i could make some errors and do a not good feeback =( by my fault, can you update code to last rev?

[alexluana]

bump pls

[divisorxx]

A little bump

[Ambal]

SilverIce does not have an internet access for now, some ISP troubles. I'll try to update this patch if I'll find some free time in nearest days

[SilverIce]

Both branches updated to last rev.

Ambal is right, better to test branch2, it filters player's relocations in much effective way

[kero99]

Patch format for anyone that want to test too =): http://pastebin.com/vPa6XLrw

Im testing at the moment at 2000+...

[SilverIce]

Thanks, i will wait for your test results

Second branch updated, AI relocation notifications optimized little more

[Ambal]

Any test results on this patch? Anyone? Guys, this piece of code is very important for us and will bring massive bonuses to core performance, why noone interested in submitting feedback?

[vladex]

ill try testing it tommorow when i get back home.

[Undergarun]

Kero99 and me are testing the second branch in a 2000+ pl realm at the moment without troubles.

Appears to be stabled but we'll continue with the test for a while...

[kero99]

Hi all, here my feedback in about 1 week testing with default patch config:

Visibility.RelocationLoverLimit = 10

Visibility.AIRelocationNotifyDelay = 1000

And:

Visibility.GroupMode = 0

Visibility.Distance.Continents = 90

Visibility.Distance.Instances = 120

Visibility.Distance.BGArenas = 180

Visibility.Distance.Object = 100

Visibility.Distance.InFlight = 100

Visibility.Distance.Grey.Unit = 1

Visibility.Distance.Grey.Object = 10

MTMaps Actived? Not active for good CPU testing

Clean Core? No, some patches

Playerlimit to testing: 2000

Core Server (DB in other server): Intel Bi Xeon X5650 2 x 6 x 2 x 2.66+GHz - 24GB DDR3 ECC RAM - Ubuntu Server 9.10

Result:

HTOP CPU Usage with 2000 Players WITHOUT Visibility and Reloc simplified: 95 - 100% -> ~98%

HTOP CPU Usage with 2000 Players WITH Visibility and Reloc simplified: 75 - 98% -> ~87%

Max gain of CPU ~20%

Standard gain: ~11%

Minimal gain: ~2%

Some problem in gameplay detected by testers? No

Some problem in stability? No

Conclusion: The patch is really good and help soo much in CPU Usage

Best regards

[Ambal]

Nice results, kero99, thank you very much Is patch in branch 1 can be applied for tests? We just want to see the difference in CPU usage between these two to decide what parts might be improved.

Cheers

[kero99]

Ok, i will apply branch 1 at weekend and test some days =)

Best regards

[Ambal]

Any test results for the patch from branch 1?

[Undergarun]

Hi ambal. Kero99 and me are now testing parallelizing client packet processing with mtmaps + TBB concurrent vectors + SQLAsyntask patch (implemented in [10865]) and maybe today or tomorrow we are going to start the branch 1 tests.

Since today we think that branch 2 is enought good! (Stable and increase performance as u can see in the kero99's feedback).

Kero99 will post the feedback from branch 1 as soon as posible.

Thank you 4 your work, we apreciate it!

[Ambal]

Since today we think that branch 2 is enought good

Thank you, Undergarun, for making tests. We need numbers to compare performance of both patches before we decide if branch 2 can be improved

[kero99]

Sorry for reply later, very busy week =(... i try yesterday to put in the production server put old_way branch not compile in linux: http://pastebin.com/3fbQM02v =(

(Compiling in clean silverice repo)

Best regards

[SilverIce]

Thank you very much for your tests

branch1 compilation fixed now: i hadn't too much time to fix it earlier

[Ambal]

Did anyone of you managed to test branch 1 of this patch?

[kero99]

lol, not soo much voluntaries . Ok ok, i have some free time again =), i would test and post feedback

Best regards

[kero99]

Easy mode patch for testing Branch1: http://pastebin.com/fD8iBsxh

Undergarun feedback soon..

[Ambal]

Easy mode patch for testing Branch1: http://pastebin.com/fD8iBsxh

Guys, you should test this branch: https://github.com/SilverIce/mangos/tree/relocation_old_ver2 instead of branch_1 Do not mess patches - this is very important since they differ greatly in terms of performance