[11193][performance] Visibility updates and relocations

[Undergarun]

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

The pastebin.com patch is wrong but we are testing the right one on our realms ;-)

[kero99]

Ups sorry, new easy patch for old ver 2: http://pastebin.com/icingfVt

=)

Best regards

[Undergarun]

Tests for visibility and reloc branch 2 (https://github.com/SilverIce/mangos/commits/visibility_op2_simplified)

Visibility.RelocationLoverLimit = 10

Visibility.AIRelocationNotifyDelay = 1000

Visibility.RelocationLoverLimit = 10

Visibility.AIRelocationNotifyDelay = 100

Tests for visibility and reloc branch 1 (https://github.com/SilverIce/mangos/commits/relocation_old_ver2)

Visibility.NotifyPeriod = 1000

Visibility.NotifyPeriod = 200

PD1: Test with 2000+ pl are not possible without a mtmaps enviroment.

PD2: After changing Visibility.AIRelocationNotifyDelay or Visibility.NotifyPeriod from lower value to higher... and tip .reload config CPU don't decrease from 100% of CPU use until next server reboot / crash.

[SilverIce]

Thank you for tests, Undergarun

As i can see from your post, AI notifications consumes much more CPU's power(~30%) than visibility update operations (~5% only)

PD2: After changing Visibility.AIRelocationNotifyDelay or Visibility.NotifyPeriod from lower value to higher... and tip .reload config CPU don't decrease from 100% of CPU use until next server reboot / crash.

I will take look at it. Both branches(old_way and simplified) have this problem?

[Undergarun]

PD2: After changing Visibility.AIRelocationNotifyDelay or Visibility.NotifyPeriod from lower value to higher... and tip .reload config CPU don't decrease from 100% of CPU use until next server reboot / crash.

I will take look at it. Both branches(old_way and simplified) have this problem?

Yes, problem happens with both branches

[Ambal]

Thank you, Undergarun Your test results are fantastic!

For branch_2 and new patch performance difference can be explained by the fact that for 1 sec player can run ~7 yards != 10 yards default in second patch => less updates. Plus, in branch_2 we update visibility only if player moves away from certain point - no updates for player if he is running around certain point => less updates This can be tricky but no worries, everything will be fine

Anyway, I and SilverIce have STRONG arguments about where we can improve optimizations

Cheers to everyone

[SilverIce]

We have taken the most effective parts from the both branches and merged them.

It's even possible it would bring more performance than second branch! (It's assumption only, testing required)

Please, compare:

merged branch: https://github.com/SilverIce/mangos/tree/relocation_opt_merged

with

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

[kero99]

Testing...

Easy mode patch for merged version: http://pastebin.com/YdCdvmry

[Undergarun]

There is something strange with new version, with 1280 pl online CPU is under heavy load (94%-99%) but with 1012 pl online and visibility and reloc branch 2 CPU load is about 58-64% so less performance?

I have to wait until tomorrow in order to make pics with 1000 pl online and to be able to compare with visibility and reloc branch 2 tests.

Stay tuned!

[Ambal]

Ouch, that sounds very weird =/ We'll take a look what might cause such issue, Undergarun.

[Undergarun]

My players reports some problems with new branch:

1. Meeting Stone stops working because of problems with distances.

2. Ritual of Summoning stop working, same problem as described before.

3. Great Feast (Cookers spell) you have to turn around it if you want to seee it! U.u

4. Fishing. You can't see the fishing (I don't know how to say in English) bobber (Thanks Vladimir)

[VladimirMangos]

bobber

[Ambal]

Undergarun, what is the following values in your config for variables:

1) Visibility.AIRelocationNotifyDelay

2) Visibility.RelocationLowerLimit

?

[SilverIce]

The problems that were reported by players are not such a big problem(already fixed) compared to CPU load problem, which is a mystery for me

[Undergarun]

Undergarun, what is the following values in your config for variables:

1) Visibility.AIRelocationNotifyDelay

2) Visibility.RelocationLowerLimit

?

Default Values:

Visibility.RelocationLowerLimit = 10

Visibility.AIRelocationNotifyDelay = 1000

Let me test it tomorrow more carefully because today i start tests when i have too much people online.

[Undergarun]

Less performance confirmed!

[Gotisch]

I am sorry but are those "moment in time" views not totally non informational? At least something like avg ammount of users per 1 hour and avg latency should be used, or at least some sort of load average / player average.

how is the cpu use at one specific moment saying anything?

[Ambal]

Undergarun, there is currently no issues in code which can cause such problems you have. Maybe patch was applied with errors? Please, check if part of code which initializes visibility/AI periods from config is not damaged.

Gotisch, statistics provided by Undergarun has enough precision to judge patch effect. You cannot control amount of players on you machine as well as sorts of activity they do. So there is no possibility to sit for hours and measure mid CPU usage values.

Cheers.

[Undergarun]

I am sorry but are those "moment in time" views not totally non informational? At least something like avg ammount of users per 1 hour and avg latency should be used, or at least some sort of load average / player average.

how is the cpu use at one specific moment saying anything?

I can't keep exactly 1000 players online for one hour and take 3600 pics for really accurate information so when population is around 1k and always in a non mtmaps enviroment to avoid CPU jumps i look htops values around 5-10 minutes taking lowest and highest CPU load (That means "CPU use: 71-77%"). Of course CPU load is always changing depending on everything, but you know any other way to get better information?

[Ambal]

In case there will be no reports about server instability with "visibility_op2_simplified" branch - it will get added into repo in nearest time as most simple and reliable solution out of all introduced patches.

It also has the highest performance due to the fact that there is no need to traverse through all cells on map/any containers to search for objects which need to be updated - this is the main explanation why this patch is faster even than "relocation_opt_merged" patch disregarding of all those optimizations implemented in it.

Cheers

[SilverIce]

In [11193]. Thank you very much kero99 and Undergarun