Jump to content

Raw UDP Sockets - Server Stability Note for Server Admins


Drathek

Recommended Posts

  • 2 weeks later...
  • Replies 82
  • Created
  • Last Reply
  • 4 weeks later...
Guest BubbaCrawfish

I've been using it for so long, that it's obvious that this thread, and the negativity expressed in it, is unwarranted.

Link to comment
Share on other sites

  • 1 month later...
7 hours ago, BeardO said:

The below is the output of top on my linux server. It's a cluster and I only set one of the servers to rawsocket and you can see the difference.

%CPU %MEM TIME+ COMMAND
116.2 15.9 7:42.52 ShooterGameServ
20.6 3.9 876:06.13 7DaysToDieServe
17.3 17.0 30:51.56 ShooterGameServ
13.9 18.3 24:14.76 ShooterGameServ

I dont get yo upoint ,sometimes the ark servers doi sit at 100+% of cpu and a restart fixes them.  I find it happens about 50% of the time when I do a destroy wild dinos.

 

Restart it and see what happens

Link to comment
Share on other sites

57 minutes ago, ONEadmin said:

I dont get yo upoint ,sometimes the ark servers doi sit at 100+% of cpu and a restart fixes them.  I find it happens about 50% of the time when I do a destroy wild dinos.

 

Restart it and see what happens

Was no point, just posting that adding the raw socket option increases CPU usage.  I tested it by turning on on another of the servers and saw the same 110-120% usage on that server.  Restart had no effect, removing option and restarting dropped it back to ~20%.  Not a big deal to me as that's core usage and I have 8 cores, just posting what I saw.

Link to comment
Share on other sites

I see the same result on my windows servers as well.

Using TCP P2P, i have an overall CPU load of about 20 - 30 %  (3 Servers / About 30 players).

On ?RawSockets, i have a CPU load of 60 - 80%

 

Even when the servers are in sleep mode and only use 200 - 300MB RAM, the CPU usage is still there.  Something seems to be programmed really wrong :)

 

Regarding Server fps:  Its about the same with TCP and UDP on the same player counts, even if there is a different CPU usage.

Regarding Laggs/Rubberbanding:   

- Bronto farming  ---> Absolutely the same laggs (server fps goes down to 2.5 for a second)

- Auto save laggs --> The same

- General lags --> I don't know because we never had laggs on TCP

 

In Conclusion:

Doesn't help anything...only generates a higher electricity bill because of the higher CPU usage

 

@Jeremy Stieglitz

Needs to be reworked again :) Please reduce the CPU load on idle and sleep mode.  

 

Link to comment
Share on other sites

Ok, yeah I am seeing this too, it never used to do this.  I notice also on server startup it is peaking at 200% cpu usage adn then falling back down to103% with no players, so are they running a second thread that is polling the network for the raw sockets???  that would use 100% cpu and explain the 200% at startup instead of the old behaviour of ~100% at startup and falling to 3% when no player online. 

Yeah, I don't want a core maxed out at 100% all the time for no reason.

For the devs info, mine is running on linux with an i7 cpu.  I know in the past there have been some issues that only happen on one os and not another so maybe that is happening here?

Link to comment
Share on other sites

Same high cpu usage here,  using i7 7700k 4.8ghz , 32gb RAM + 512ssd + Windows 10 ,  8 cores (4 cores + 4 Hypertheading.) only running 1 server.

width rawsockets, all cores goes to 30/50%% width very  high and quick peaks (70%-90%) to all cores, no people playing,  and no  detection <   lag raiding in game and not  better performance widh big bases.

Adding new parameter -nonetthreading to starting command line, cpu goes to normal ussage (3% - 4 %)

Link to comment
Share on other sites

10 hours ago, docjones said:

Same high cpu usage here,  using i7 7700k 4.8ghz , 32gb RAM + 512ssd + Windows 10 ,  8 cores (4 cores + 4 Hypertheading.) only running 1 server.

width rawsockets, all cores goes to 30/50%% width very  high and quick peaks (70%-90%) to all cores, no people playing,  and no  detection <   lag raiding in game and not  better performance widh big bases.

Adding new parameter -nonetthreading to starting command line, cpu goes to normal ussage (3% - 4 %)


is that the correct spelling? 

Either way I tried -nonetthreading and -nonethreading with no change in CPU usage on Linux server.

Link to comment
Share on other sites

Guest BubbaCrawfish

I do have to say this is an interesting scenario... I also have added the nonnetthreading option, and had it drop from a solid 100%, to 10-12%.

I'll be watching the performance of the server with anticipation.

Link to comment
Share on other sites

Found some massive problems with bRawSocket!

Ragnarok has problems to create the savefiles when using bRawSockets. Its creating .tmp files instead of .ark files. 

 

So if you do not want to loose data, i recommend you to go back to TCP (Steam P2P). Anyway there is zero performance improvement by using UDP.  The autosaves generally runs better with Steam P2P...so less lagg by using TCP. 

 

I will give it another try when its really fixed...which is absolutely not the case at the moment. More CPU load, no performance improvements and problems with save files. 

Link to comment
Share on other sites

7 hours ago, Toni said:

Found some massive problems with bRawSocket!

Ragnarok has problems to create the savefiles when using bRawSockets. Its creating .tmp files instead of .ark files. 

 

So if you do not want to loose data, i recommend you to go back to TCP (Steam P2P). Anyway there is zero performance improvement by using UDP.  The autosaves generally runs better with Steam P2P...so less lagg by using TCP. 

 

I will give it another try when its really fixed...which is absolutely not the case at the moment. More CPU load, no performance improvements and problems with save files. 

Thanks for the heads up

Link to comment
Share on other sites

Thats the last patch notes for Current Version: v271.17

- Defaulted dedicated server ?bRawSockets mode to not use threaded networking, seemed to generally be a net performance loss in our analysis. Can still be forcefully enabled with -forcenetthreading

But something changed widh this update.

My server cpu % ussage increased again, using rawsockets.  now param -nonetthreading don't have any effect.
No problems saving my big file server ragnarok.ark (230Mb.).
If this patch not use threaded networking, then  cpu % increase. but better network and minor lag ?
Then is better for lag (big bases) using  rawsockets  or p2p?

 

Link to comment
Share on other sites

11 minutes ago, docjones said:

Thats the last patch notes for Current Version: v271.17

- Defaulted dedicated server ?bRawSockets mode to not use threaded networking, seemed to generally be a net performance loss in our analysis. Can still be forcefully enabled with -forcenetthreading

But something changed widh this update.

My server cpu % ussage increased again, using rawsockets.  now param -nonetthreading don't have any effect.
No problems saving my big file server ragnarok.ark (230Mb.).
If this patch not use threaded networking, then  cpu % increase. but better network and minor lag ?
Then is better for lag (big bases) using  rawsockets  or p2p?

 

The saving problem is not reproducable, its coming randomly.  An ark server alsways first creates a .tmp and then moves it to a .ark.  In my case about each 10th time it couldn't make the .ark file and leaved the .tmp file. 

 My observations about UDP:

- Autosave gives a much bigger lagg than with Steam P2P

- Big bases are loading in slower with UDP

- Generally (Gameplay) no differences found.   Not better, not bader...

- Uses more CPU Power than Steam P2P (But has the same ingame server performance)

 

So in conclusion, still the same --> Its just heating your office/serverroom with the unneeded CPU usage it generates. No benefits by using it.

Link to comment
Share on other sites

Is there an idiots guide for how to use this? :)

I setup port forwarding on my router for what I *think* is needed for the ports (7778,7779,27051,27052 tcp and udp) but when I start the server up listening on those (?QueryPort=27051 and ?Port=7778) , I can't connect. If I remove ?bRawSockets I can connect just fine.

Documentation in the wiki is lacking.

Link to comment
Share on other sites

Newest example:

Many admins with the S+ Mod have corrupted saves and can't startup their servers. (Exactly on the map where i said bRawSocket has a Problem with the Save Files -> Ragnarok).

I don't have this crashes/corrupted files on my servers. (Maybe because i gone back to TCP)

So there is a big chance that this problems are caused by bRawSockets ?

Unfortunately no of the admins with crash servers is answering me the question if they are using bRawSockets or not...Each one is posting the Crash stack instead of answering questions which can help to find the cause.  

 

Link to comment
Share on other sites

Hi,

this is not really the subject, but this may be caused by their ""fixes"", have you any stability problem caused by the auto-decay of abandoned structures? I've noticed multiple server crashs and huge memory leaks (up to 16GB instead of 7/8) since last weeks caused by the auto-destruction of abandoned structures (much more than before), and a general deterioration of the game server perfs since official release; disconnecting, crashs, memory leak, freeze, rubberbanding 

Same thing with destroywilddinos command, heavier than ever, can cause 20 minuts+ freeze or even crash

-Linux
-TCP P2P
-TheIsland/TheCenter
-No Mods
-.ark files much smaller than a few months ago
-Dedicated server, no hardware problem/300mb stable connection
-Correct or native rates/parameters

 

...Anyone else?

Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.


×
×
  • Create New...