We at Dead Men Racing are having nightmare connectivity problems and would like to call upon the power of fellow hosters to help us sort out the problems.

We do not suspect our hardware or network to be at fault. We run an absolute beast of a server hosted in Hi-Velocity's London data centre.

Here's an example from last night's racing ...

The server had 16 racers on. Lag was obvious with the usual problems of warping cars and 'lag tags' above some cars. The behaviour of the Latency/Lag bars (or as we call it - the ping wiggler) was not quite as documented. We were connected by Ventrilo so we were able to keep each other informed of what we were seeing. The most common anomaly was when three or four racers would report (at the same time) that their bar was rising and that they were probably about to drop. What then happened was one of those would drop and the rest would quickly restore to normal. The link above seems to say that this shouldn't happen. Sometimes all the people with rising bars would drop - even though they're on different ISPs.

We've been running Pingplotter traces which show no real issues so we're stumped.

Is there any thing we could try to help solve these issues?

I should also add that we also run a CounterStrike server and the players on that server are seeing no lag at all.

TIA

Could this be a clue?

http://www.lfsforum.net/showthread.php?p=427339#post427339

Does LFS have known issues with dual processor systems? We run a dual Opteron 265 based system so, if it does, this could be a likely cause.

Quote from OldBloke :

We do not suspect our hardware or network to be at fault. We run an absolute beast of a server hosted in Hi-Velocity's London data centre.

These aren't mutually exclusive, and having a "beast" of a box isn't a good argument for trouble.

As for dual core CPU's, I've not noticed any trouble specifically whilst testing. You might want to poke SamH / mkserve / the UKCT team about this, as they've just migrated to a new DC CPU'ed box, so they may have started to notice this especially given the sheer number of racers they cope with.

The trouble with debugging, using ping data, is that they're not streams of data in the true sense - there is a distinct separator between where the data stops and starts.

It's probably worth throwing a network debugger onto your offending machine and logging the data and trying to analyse it and spot a pattern. If you find it occurs at certain times of the day there's something your data goes through which is overloaded. If it occurs all the time, then it's more likely to be within your hosts network - be it your box, or something on the route out. If it doesn't occur on connections to specific ISPs, then it's more likely (but not certain as they should have multiple gateways) to be an issue outside of their network.

Talk about timing! LFS hasn't, as far as we've seen, any issues with our dual-core. However, last night for no reason we could figure and for the first time ever, our BF2 server *suddenly* started showing lag problems. We shifted its process usage on to one core and the problem was instantly resolved.

I don't know if this will solve it, but try doing the same with your LFS server

PS: If you figure out a way to set an individual core or processor as the one an application process uses *by default*, each time it's launched, I'd love to hear about it.

ForceCore ftw.
http://www.kalme.de/index.php? ... w&id=44&Itemid=35

Certain releases of Windows have an /AFFINITY switch for cmd, which you can use to do the same thing. I *believe* it's x64 only, but don't quote me on that.

Quote from the_angry_angel :

ForceCore ftw.
http://www.kalme.de/index.php? ... w&id=44&Itemid=35

Certain releases of Windows have an /AFFINITY switch for cmd, which you can use to do the same thing. I *believe* it's x64 only, but don't quote me on that.

you can easily set the affinity via the task manager Processes tab also. and that works fine with windows 32 bit

Using the taskmanager, it won't persist - which is the problem here.

ForceCore allows you to script it and just let your server manage itself (which any good server should).

Using the START command:

http://www.techenclave.com/for ... nity-made-easy-67317.html

Quote from SamH :

Talk about timing! LFS hasn't, as far as we've seen, any issues with our dual-core. However, last night for no reason we could figure and for the first time ever, our BF2 server *suddenly* started showing lag problems. We shifted its process usage on to one core and the problem was instantly resolved.

I don't know if this will solve it, but try doing the same with your LFS server

PS: If you figure out a way to set an individual core or processor as the one an application process uses *by default*, each time it's launched, I'd love to hear about it.

Thanks for all the info, guys.

I used imagecfg.exe (from the Windows resource kit for W2K3) to set lfs.exe running on a single cpu and whaddya know, no mass disconnects. I realise that one night's testing is hardly conclusive so I'll give it to the end of the week then revert back to no affinity and see if the problem returns.

TAA - Will give ForceCore a look. Thanks for the headsup.

Quote from the_angry_angel :

Using the taskmanager, it won't persist - which is the problem here.

ForceCore allows you to script it and just let your server manage itself (which any good server should).

good point. i was just thinking short term .