Server spike issues on VDS

[Mr.Tickles]

Greetings,
I am currently using an unmanaged VDS with 4 HT virtual cores and 7GB memory. On windows server 2012 located in chicago.

For quite a while now I have been running three public TF2 servers (24 slots a piece), a private TF2 server (18 slots), and two killing floor servers (6 slots a piece).

I have been running all this just fine for around 3-4 months without issue. This past week our public TF2 servers have suddenly started experiencing stuttering issues every few minutes. My first thought was that it was a resource issue so I checked CPU and Memory. Cpu was doing just fine at around 48% but memory was high at 95% usage. I figured what was happening was we were having resource spikes on memory causing stuttering issues. So a couple night ago I added an extra GB to give us some breathing room.

Sadly this did not resolve the issue. I know that source servers have issues with HT cores due to having a stuttering issue like this one before. I had resolved it by giving each of the TF2 servers one of the four cores via assigning their affinity. Thinking that perhaps after the move (which was lossless) it had somehow changed the affinity back to standard. This was not the case.

After watching the servers via HLSW (rcon program) I noticed that the connection spikes on them one at a time. Not together and no noticeable pattern either. I am not certain what would have changed to result in server issues.

My only thoughts are that perhaps its a source update that came out recently that is causing the issue but I have no idea what to do to address the problem. Any ideas?

[soja]

Add me on steam: http://steamcommunity.com/id/soja992

Post some server IPs to check out, I have had this issue before and I think I know what is causing it. I would like to play on a server to confirm though.

[Mr.Tickles]

Server ips are tf2app.com:27015 badwater tf2app.com:27016 nucleus tf2app.com:27017 hightower

[kraze]

There was an update on the 16 so it is possible that update broke or caused issues with SM/MM or any of your plugins. Have you gone through your addons and updated them to the latest stable version?

Also when you see this stuttering are you seeing any choke/loss in net_graph 4 or any ping spikes?

[soja]

I have played on it for about an hour now and I noticed one long spike out of that gametime, which without repeating it i could not explain.

how are you determining lag spikes? Strictly HLSW or are you watching net_graph 3/4 in-game?

You can fix the spikes when coming out of spawn with net_splitpacket_maxrate 100000. This will also result in a smoother gameplay experience overall.

Tested on your hightower server 20+ for the course of my time.

[Mr.Tickles]

To be honest I did not think to use netgraph 3/4 at all. Mostly off the reports of my regulars and from what I what I have seen in HLSW.


I'll check on the plugins themselves but since this has been happening since before the update yesterday I imagine that there might be something else causing the issue. No harm in checking though.

As for the net_splitpacket_maxrate 100000 setting I think I already tweaked that based on other threads I found in the NFO knowledge base for stutter issues. I'll double check this as well.

I'll report back once I have checked all this.

[soja]

Playing on the server felt like it was at the default 15,000. Check when your server is running by running "rcon net_splitpacket_maxrate" in the console, it will reply with the current applied value.

[Mr.Tickles]

Playing on the server felt like it was at the default 15,000. Check when your server is running by running "rcon net_splitpacket_maxrate" in the console, it will reply with the current applied value.

I'll be dipped. Ran it from my phone application and it is on 15000. I'll alter this when I get a chance to tonight and see if that helps. Not certain why it suddenly started a problem then since it has been running on this I imagine since startup.

[soja]

It wouldn't be effecting HLSW showing the server spiking, but like I said, providers an overall better play experience in-game.

[kraze]

In my experience those red bars you're seeing in HLSW are just a ping graph. Meaning if you were to personally see a spike in that graph it could represent an issue or routing problem between you and the server. That doesn't mean it doesn't represent a problem, but take it with a grain of salt.

Did you ever get a chance to check net_graph 4? Any choke/loss? I'd also recommend you test the server with addons disabled for a short period just so you can see if the issue is happening or not.

[Mr.Tickles]

In my experience those red bars you're seeing in HLSW are just a ping graph. Meaning if you were to personally see a spike in that graph it could represent an issue or routing problem between you and the server. That doesn't mean it doesn't represent a problem, but take it with a grain of salt.

Did you ever get a chance to check net_graph 4? Any choke/loss? I'd also recommend you test the server with addons disabled for a short period just so you can see if the issue is happening or not.

Yeah, certainly not taking HlSW connections at face value as I know that when both myself and my dev have HLSW open on the same local network it causes both us to get red bar connections every 15 seconds or so.

However, as far as the spikes are concerned it does not completely top out it just goes up quite a bit. We can match the spikes with the exact moment a player notes it in game.

As for the ping graph I will be testing that out tonight when I get home and see if I notice anything happening there.

[Edge100x]

Fixing net_splitpacket_maxrate should definitely make a difference. It leads to exactly this problem with spiking/skipping.

[Mr.Tickles]

Modified the splitpacket and it resolved all the problems as far as I can tell. Thank you all so much for your help on this. NFO rocks at support both the staff and the forum community!

[Edge100x]

I'm glad it was resolved! soja, thanks for singling out net_splitpacket_maxrate. It's also in our performance troubleshooting guide, but it's surrounded by other variables to look at, so it's easy to miss.

[soja]

Glad to hear it! NFO quality at its best managed had 100,000 set by default, even though it raises cpu usage