Solved Block lag, but TPS at 20

[DaNoobFR]

Hi everyone,

I've spent the afternoon trying to track down a performance problem on my server. I need your help.

• What OS are you running?

Debian 7

• Are you using minecraft hosting provider or a dedicated/local server not designed for minecraft?

I'm using a VPS

• What version of FTB are you using? (Dont just say latest tell us the version)

Direwolf20 1.0.23

• Did you add any mods to the server pack?

No

• What are your server specs?

AMD Opteron(tm) Processor 4386 @ 3.1Ghz
4GB RAM
100Mbps for the server, 100/50Mbps for my home, 16ms ping between the two.

With no players connected, server is running at 100% CPU (one thread), 1.6GB RAM used.

• A good detailed paragraph about the problem

So as I said in the title, the main symptom is block lag. MFR machines are slow, it takes seconds to pick up items, blocks reappear, etc... The weird thing is '/forge tps' always gives me around 27ms tick time overall. Opis on the other hand shows an update time that goes from 50 to 250ms.

There's only one other active player on this server and I've broken his chunk loaders. So the problem must be in my base : I couldn't find a problematic chunk though, or a tile entity that would eat up too many resources. The only thing that stands out in Opis is the Handler 'ThermalExpansion Grid Tick Handler' at 24525us.

A bit of googling pointed me to itemducts/fluiducts. I replaced most of them with a combination of AE and tesseracts. Didn't change a thing. So now, I'm wondering if it could come from energy conduits. And if so, I'm in a world of pain : they are everywhere...

Guys, I'm out of ideas. Any help would be much appreciated.

Edit: I switched from OpenJDK to Oracle Java, same CPU usage, memory usage a bit higher, but the problem is still there...

 

[Johnson]

If Opis is showing 250ms, you are getting HUGE TPS loss. Anything above 50ms will be a tps loss. When you look at Opis, what is the source of your large ms time? Check tile entities specifically and then check the handlers tab and check your GridTickHandler time.

 

[DaNoobFR]

Thanks for taking the time to answer. I don't want to be rude but could you please take a little more time to read the problem description, before answering :
I know 250ms in Opis is huge, the weird thing I pointed is '/forge tps' always reported around 27ms. Why would forge report 20 TPS when there is clearly not?
Thanks for suggesting to check the Handlers. I did, as you can see in my original post : GridTickHandler was at 24ms, according to Opis.

Anyway, I think I found the origin(s) of the problem. There was an ender-quarry in the nether, with backstuffed itemducts. And a skeleton farm which didn't pick items up. Broke them both, whipped the newbie (just kidding), and we are back to 15ms overall.

Any idea what it would take to make itemducts appear in the tile entities listing of Opis? That would help _a lot_ in that situation.

 

[Johnson]

They won't appear in Opis, but here is a server side mod than can help. To use it, run the following commands, /pgobbler scan 10 and then /pgobbler purge 10. What this does is scan all itemducts that are stuffed with 10 or more stacks of items. It will give you the location and how many stacks, then the purge cleans it up. http://www.curse.com/mc-mods/minecraft/pgobbler

Hope that part helps at least.

 

[DaNoobFR]

Thanks. That should help a lot.

 

[UnionCraft]

They won't appear in Opis, but here is a server side mod than can help. To use it, run the following commands, /pgobbler scan 10 and then /pgobbler purge 10. What this does is scan all itemducts that are stuffed with 10 or more stacks of items. It will give you the location and how many stacks, then the purge cleans it up. http://www.curse.com/mc-mods/minecraft/pgobbler

Hope that part helps at least.

Just to note, The pipegobbler doesn't resolve the TE Gridhandler issues, if your gridhandler is still high its going to drive TPS down and gradually get worse until the server is unplayable, the issue is caused by items moving across chunk boundaries via TE ducts. Our Gridhandler was at 16000+us and climbing, the lag spikes and TPS basically killed the server, since we removed ducts the issue has gone completely. That was 4 weeks ago. Its now at 0.050us and doesn't generally move if we have 5 players on or 50.
For some reason it doesn't seem to matter if the chunks are loaded or not, if you have ducts anywhere in your world, the Gridhandler will spiral out of control.

We had reported this issue to the Cofh team, it was ignored.

 

[DaNoobFR]

I came to the same conclusion : all players on my server have been advised to limit TE pipes to small networks inside a single chunk.
For everything else, they are to use tesseracts and AE.

I really hope the COFH team will take this problem into account.