Alien Swarm + BitTorrent = Server Lag?

[Ronin Storm]

Just got AS 1.3 and the Telic campaign installed. Wandered onto server #2 around 21:00 last night to join in, but the performance was really bad -- serious warpage. Killer-mk1 commented that it had only gotten bad since I arrived. Not sure if that was in jest, but I took it not to be so I logged off and went to check the server from the command line. A load of around 0.5 didn't seem serious so that didn't seem to be the issue.

Then I noticed it... I was furiously downloading Anime via Azureus (my BitTorrent client). Now that'd explain why my performance was screwed but it set me to wondering if that could also explain poor server performance.

Now, I know (I really do know) that this is pretty unlikely as the server should simply treat "no update" from me as "he's done nothing" but what if Alien Swarm or UT2004 has some sort of control connection that I was throttling with my BitTorrent that really was leading to poor server performance (though likely not as bad as mine!)

This totally inconceivable?

 

[Pestcontrol]

Yes, Unreal simply doesn't work that way. It's all asynchronous, so no degree of lag on one client can affect any of the others.

And the server is only lagging when everyone experiences the lag. If it's not lagging for one person, it can't be the server.

About the cpu usage, it can be rather spikey for AlienSwarm, one moment it's just the six of you walking, the next you're fighting a dozen aliens, but yeah, if none of the other servers are being used, that won't be the problem.

I've found lag to be a strange thing psychologically. People will blame anything and everything for it, search for odd explanations, while most of the time it's just random. And at the end of december with lots of people taking time off, i would expect the internet to be more busy than usual, and lag can always occur randomly.

 

[Haven]

A quick sanity check before I look into this further. Why do we have four processes for the unreal tournament server running ? Is it two per server instance or have we actually got four servers running ?

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
1231 pestcont 31 16 306m 151m 5432 S 0.3 15.2 421:37.93 ucc-bin
1450 pestcont 31 16 306m 151m 5432 S 0.0 15.2 0:00.45 ucc-bin
1235 pestcont 31 16 278m 126m 5412 S 0.0 12.8 142:16.18 ucc-bin
1451 pestcont 31 16 278m 126m 5412 S 0.0 12.8 0:00.33 ucc-bin
13711 killcraz 29 14 82388 65m 7636 R 0.0 6.5 0:57.04 hlds_i686
13713 killcraz 29 14 82388 65m 7636 S 0.0 6.5 0:00.00 hlds_i686
13714 killcraz 29 14 82388 65m 7636 S 0.0 6.5 0:00.41 hlds_i686

Between half life and Unreal Tournament we are using 75.5% of the system memory ... Can you check that you actually intend to run all these UT processes pestcontrol ? If anyone see's KC around then it might be worth getting him to double check what he's running as well. Bearing in mind that the above stats were taken when the server was NOT loaded I'm sure you can appreciate that the situation is worse when it is. At this stage freeing up some memory so the system doesn't thrash the swap space under load is probably our first priority - this will mean dropping at least one and possibly two games servers ( an instance of AS, CS or NS depending on who cries loudest )

 

[KillCrazy]

I'm only running one instance of Natural-Selection, but it always shows up as three instances for some reason. They all dissapear when I shut the NS server down.

EDIT: I don't know what Hotstuff plans to do with NS as he wanted to start to try and seed the server again, but I have not spoken with him recently. If need be I will shut it down until we know what the deal is.

 

[Pestcontrol]

Haven: This is normal, each ut2004 instance has two threads. Doesn't mean there's double memory usage, 151MB and 126MB is what the servers use respectively.

Only the PID's in that list look strangely juggled. Usually the two threads of the ut2004 server have pid's only a few numbers apart, like the half-life process has.

 

[Haven]

hmmm I wonder if we can show shared memory usage effectively using top ... that would be useful info to have to get a more accurate picture.