Difference between revisions of "Profiler"

Latest revision as of 09:27, 27 October 2019

Profiler

Profiler is a built in mod, activated via "profiler.load = true" in minetest.conf file.

Usage: "/profiler save" by a server admin. This makes a text file "profiler date-time.txt" to check out.

The file I (Thailand) got, is from 11.09.2019

The file contains summary and single part from all activated mods and may help to target lag source.

example

more_chests:                                            
 - on_joinplayer[1] ..................................

with numbers for

min µs
max µs (leads to peaks, used for rank peaks)
avg µs (lead to avg lag, used for rank B)
min %
max %
avg % (lead to avg lag, used for rank A)

avg : mean time making work for cpu
peak : short but heavy usage, maybe feel lag

results

summary

The most avg work, mean all time through the server is on is made by mods:

advtrains (32 %)
and + messecons (13 %), mean sum cpu for 2 mods 45%
+ pipeworks (10 %) = sum top 4 with 55%
+ technic ( 9 %) = sum top 4 with 65%

check the peaks out of result pages ...

conclusion

There are main just 3 mods there behaviour have to be checks / reduces to get lag down.

advtrains, with then also japan train
messecons
pipeworks

all others are minor effect to server.

Author

Thailand, known as Festus1965 in Forum, and Thomas or Explorer on most other servers.

I played around with profiler since at least 2 years and successful reduced lag on my server basing on profiler results.
There is a change possible in pipeworks reducing that mod about 50% of its impact.
At advtrains I have no glue, as I dont use it, as I know its impact on servers, or pick up all the not used trains of not ready lines.

@@ Line 28: / Line 28: @@
 === results ===
-To read that you need to find out the most top 5 or 10 avg, as there where this time (11.09.2019):
+* [[Profiler-23102019|Profiler 23.10.2019]]
-* '''[[advtrains]] - 35.6 % avg (rank 1)''' and just 40.4% peak of peak max [[messecons]] (rank 5)
+* [[Profiler-11092019|Profiler 11.09.2019]]
-* [[pipeworks]] - 17,9 % avg (rank 2), depend on avg us rank 3, and its peaks 53 % strength of [[messecons]] peak = rank 2
-* '''[[messecons]]''' - 12.7 % avg (rank 3), '''but depend on avg us rank 1, top peak rank 1'''
-* [[technic]] - 4.6 % avg (rank 4), depend on avg us rank 4 also, peaks just 35% as of [[messecons]], rank 7
-* [[advtrains_train_japan]] - 4.6 % avg (rank 4), depend on avg us just rank 7
-* default - 4.4% avg (rank 6) ... where I stop to check, as above mods make together 75% of all cpu work ...
-* ... (a view others as asked for, or to compare also transport mods)
-* pamdorabox_costum - 2.9 % avg (rank 7)
-* advtrains_train_steam - 2.4 % avg (rank 9)
-* advtrains_train_industrial - 1.7 % avg (rank 10)
-* ambience - 0.9 % avg (rank 13)
-* signs_lib - 0.4 % avg (rank 16)
-* elevator 0.1 % avg (rank 30)
-* travelnet - 0 (rank 45) (rank 45 is all last rank, mean nothing to work itself)
-" ? carts is not listed (but from my server I know that is also very low, never had to mind abouit even with 25.000 nodes long cart system)
 === summary ===
 The most avg work, mean all time through the server is on is made by mods:
-* [[advtrains]] (35-40%)
+* '''[[advtrains]]''' (32 %)
-* + [[pipeworks]] (18%) = sum 53-58%
+* and + [[messecons]] (13 %), mean sum cpu for 2 mods 45%
-* and + [[messecons]] 13%, mean sum cpu for 3 mods 66-71%
+* + [[pipeworks]] (10 %) = sum top 4 with 55%
+* + [[technic]] ( 9 %) = sum top 4 with 65%
-The main hard peaks are generated out from
-* [[messecons]] peaks, counted here as 100% as most strong peaks
-* [[pipeworks]] peaks, at least strong 50% strength of [[messecons]]
-* [[advtrtains]] peaks, which are still 40% as strong like the [[messecons]] peaks
-* [[technic]] peaks, have also 35% the strength like [[messecons]] peaks
-* ambience then follow with 24% hard peaks compare to [[messecons]]
+check the peaks out of result pages ...
 === conclusion ===
-There are main just 3 mods there behaviour have to be checks / reduces to get lag down.
+There are main just 3 mods there behaviour have to be checks / reduces to get [[lag]] down.
 * [[advtrains]], with then also japan train
 * [[messecons]]
 * [[pipeworks]]
 all others are minor effect to server.
 === Author ===
-Thailand, known as Festus1965 in Forum, and Thomas or Explorer on most other servers. I played around with profiler since at least 2 years and successful reduced [[lag]] on my server basing on profiler results.
+[[Thailand]], known as Festus1965 in Forum, and Thomas or Explorer on most other servers.
-There is a change possible in [[pipeworks]] reducing that mod about 50% of its impact. At advtrains I have no glue, as I dont use it, as I know its impact on servers.
+* I played around with profiler since at least 2 years and successful reduced [[lag]] on my server basing on profiler results.
+* There is a change possible in [[pipeworks]] reducing that mod about 50% of its impact.
+* At advtrains I have no glue, as I dont use it, as I know its impact on servers, or pick up all the not used trains of not ready lines.
 [[Category:Server]]