xyvyx

So if this a Netty issue, I wonder what the common thread is.... network hardware? If so, swapping in a new NIC could be an easily solution... if that's what it turns out to be.

My issue was resolved by the NOGUI switch and I don't remember ever seeing any netty IO methods eating enough CPU time to show up on the VisualVM radar.

Vanilla 1.72
Java V.7 update 45
Windows 8.1 x64

My network card is either an Intel or an integrated Marvel chip, I'll check when I get home. If this issue affects both Linux and Windows, that makes me think it's Netty + network hardware related.

@Hunter Parkhurst - yeah, add the NOGUI option and see what happens.

I think it's related to how it draws the "activity graph", which stopped working w/ this version also. Even though it's obfuscated, we could run VisualVM and see if it's the same object/method calls eating up CPU. Granted, in my case, it was all the GUI...

are you running the server with our without the server GUI?

I've got a vanilla 1.7.2 server on Windows 8.1 x64. Core i5 machine w/ 16GB of ram and MC sitting on a fast SSD. The java process would consume 25-50% CPU (it was lower when the world was first created) when idle and more when a few clients were connected.

The higher the CPU load, the more "can't keep up" messages. Sometimes it would spike to 80-100% CPU, but it'd typically drop again after a few minutes.

Last night I added the NOGUI option to my startup command... wow... night/day. When I looked at it this morning, the machine was idling at 1-2% CPU. As users played last night, I still saw some of these warnings in the log, but I think it was only when they were exploring new areas.

Given that the GUI for the windows EXE doesn't even work, this isn't too surprising. Could be related to it trying to draw the activity/load graph... which off-hand, I can't remember if that was visible w/ 1.7.2?

It's possible this started with 1.6.x and was complicating the other "zombie pathfinding lag" issue.

curious, are you running the server with or without the GUI?

So do you think this is related to the "nogui" issue you posted after this one? It looks like MC-39078 was deleted? but I saw the cached version where you said setting the nogui reduced the CPU down to 1-2%. I'm having similar issues... CPU usage on an idle server (no clients, hour+ after last person disco'd) hovers from 25 - 45% on a 4-core 3.4ghz i5 processor. CPU load appears evenly distributed across all 4 cores. Running VisualVM shows the largest time consumer is some javax.swing.timer call (or DoPostEvents or something like that, don't have it in front of me)

Console appears the same for me. Sending a /stop command still appeared to work & clients can connect just fine. Server's log file wasn't being written to, so maybe that changed as well... not sure yet.