Chunk Load Thrashing
Opened this issue ยท 6 comments
GreenMarine commented on May 18, 2019, 5:49 AM UTC:
Modpack version
1.64a
Can the issue be reproduced?
Yes
Logs
https://pastebin.com/mnS2zeyc
Issue
Hello! I'm back. After talking to the EnderIO guys here:
Which was a result of this discussion here:
We determined that a likely cause of the EnderIO network lag is chunk load thrashing. Apparently when chunks are loaded and unloaded, the network rebuilds itself. I ran tellme with a monitor on chunk loading and got a fair amount of rapid chunk load / unloads.
Some of these can be seen here: https://pastebin.com/mnS2zeyc
and a lot more here: https://pastebin.com/2ByXAtgv
Logs were taken with:
tellme track enable all-dims chunk-load
tellme track enable all-dims chunk-unload
This unloading / loading cycles at a high rate and happens even if users are offline.
Loaded dims:
[tellme]: DIM 0: overworld [ 358 loaded chunks, 118 loaded entities, 0 players]
[tellme]: DIM -343800852: Spectre [ 0 loaded chunks, 0 loaded entities, 0 players]
[tellme]: DIM 2: Storage Cell [ 0 loaded chunks, 0 loaded entities, 0 players]
(The loaded chunks count in dim 0 will cycle fairly quickly.)
Examining entities in the thrashing chunks indicates nothing particularly suspect.
This issue was moved by NillerMedDild from NillerMedDild/Enigmatica2Expert#1194.
NillerMedDild commented on May 18, 2019, 7:02 PM UTC:
Reopening in case someone can help here, as everyone stand to benefit from better performance :D
Ricket commented on May 22, 2019, 3:49 AM UTC:
The rapid chunk load/unloads might be the issue I opened here: MinecraftForge/MinecraftForge#5406
I was able to help relieve the issue by increasing the forgeChunkLoading.cfg dormantChunkCacheSize. Effectively making it so that these chunks are only thrashing into and out of the dormant cache, not into/out of disk.
The comments on that issue might indicate that this issue is fixed in MC 1.13.1, I'm not certain.
Ricket commented on May 22, 2019, 3:51 AM UTC:
BTW you mentioned dimensions at the end -- if you are also having dimension loading thrash then there's a similar variable in forge.cfg for dimensions: dimensionUnloadQueueDelay. You could try increasing that to make a dimension stay loaded longer, which would reduce thrash.
And of course I am not recommending specific numbers for these configs because it's very dependent on the situation and the amount of server memory (and CPU to some degree).
GreenMarine commented on May 22, 2019, 4:44 AM UTC:
The dimension trashing was due to one of our players chunkloading part of the nether - which might have then triggered this chunk load/unload problem. What's described in that issue you linked does match what we see on our server. In fact, several times when I teleported to a problematic chunk I found a cow or other entity on the edge of it. It struck me at the time, but I didn't have a model for linking that to the problem. Now I have that model.
NillerMedDild commented on May 22, 2019, 9:09 AM UTC:
I increased the dormantChunkCacheSize and set dimensionUnloadQueueDelay to 200 ticks, hopefully that'll have an effect.
GreenMarine commented on May 18, 2019, 6:10 PM UTC:
Closing this. I couldn't figure out the problem, but migrated the server to an i9. Solve it with more compute!