Server setup.
 

Hi guys, old troll, first post. I got my hands on some hardware through work and been wanting to host a server so seems like a perfect oppertunity to get started. I hope this post isn't supposed to be in the support form :)

Which is most important when it comes to server hardware, CPU or RAM?

What will put the most amount of strain on the server, EQEmu itself or the database?

I've been thinking of hosting the EQEmu server on a Dell Poweredge 860 (Dual core, 1,86 GHz, 2 GB RAM, Raid 1) and running the database on a old HP workstation (pretty weak dual core, 4 GB RAM, Raid 1). Would this setup be to prefer instead of just putting it all on the same dellserver?

Might invest in abit more RAM to the poweredge in the future.

Anyone noticed any differences in performance when it comes to OS?

Anyone tried running EQEmu on Windows Server 2008?

Still haven't decided what OS to use.

from what I know:

-there is no benefit of having Server and Db to be run off diffirent PCs - you may very well humper its performance (the DB of EQ is REALY tiny compared to what corporations are running - the whole thing is at most 100mb)

-RAM comes first, CPU 2nd

-The real drain on the server is number of open conection (aka players), and this is where UP bandwith realy matters

-Time has proven that Unix run Emu is a bit more stable and runs faster

Unfortunately the EQClient, Server, or Database make no use of multiple core processors. Like Chaos said, your biggest focus should be RAM, then CPU. Besides that you'll need a crap load of bandwidth if you plan on having even a small community.

An empty zone (dynamic or static) takes about 8MB of RAM. The database/MySQL will take about 200-300 MB of RAM. I've seen some zones get up to 70-90MB's with a couple groups in them.

I honestly would recommend 3 GBs of RAM minimum on a 32-bit OS (or 4 GB minimum on a 64-bit OS).

I know from experience that 768 kb Up is only good for about 40-50 people. That is really stretching it though due to the variances in packet size for different actions. I saw fluctuations between 300 kb UP and 700 kb UP with only 30-35 people on the server so it's very possible 40+ can cause issues with only 768.

We recently moved to a 20 Mb UP line and it's helped tremendously.

Be careful with the Dell Poweredge under Linux. If it has a PERC raid controller then you'll have piss-poor filesystem performance. Replacing the PERC card with another raid card is well worth it.

If you are not using hardware raid, change it in the bios to act as a SCSI controller Kobaz. There are no known issues with any PERC controller on Linux, as hundreds of thousands of Dell PowerEdge servers do run Linux. You most likely had a configuration issue.

Actually, some of the PERC cards are problems under Linux on Dell. The PERC 5/i was particularly notorious for having issues. There are some driver updates that were supposed to help, but they didn't always.

EDIT: I should add that not ALL perc cards are issues and not even all PERC 5/i's had issues. But there are a greater number than others. Also, I never checked if they had issues on Windoze too.

Thanks Andrew. I couldn't remember which model it was. And there were no problems with it under Windows, it was a quite nice card for the price.

EQEmu makes no use of multiple CPUs? That's not what I have heard. From last I checked EQEmu will eat up as much CPU as you can throw at it..

Although, I do recall a PEQ staff member telling me it was running on a P4 3.1GHz with 2gb of ram, and they get 150 people at peak.

230 peak ;)

EQEmu definitely will make use of multi-core CPUs. It isn't able to spread the load if any particular process between the separate cores, but it will split which cores each process loads on so it divides up pretty equally. Some zones will run on core 0 and the others will run on core 1 for dual core. Similar for a quad core I imagine.

Thanks
 

Thanks a ton for the info. Decided to go with Debian and threw in an additional 4 GB of RAM. It will use a 100/100 Mbit line so hopefully bandwith shouldn't be a problem. It is a Eurotrash server so I'm abit worried about lag and MS for US users. Time will tell, thanks for the input.

Technically, however that isn't EQEmu's doing. That's just the CPU's natural load balancing techniques.

Well, the term Linux is a bit broad these days. I've never seen any problems with PERC controllers on Redhat, CentOS, Oracle EL4, ESX, or Debian-flavor systems. All we've used at the office over the past couple years has been PowerEdge 2850s and 2950s servers running Linux. We have found bugs in the system BIOS (not to be confused with the PERC bios) which have required new releases from Dell to fix. :(

Actually, it technically is EQemu's doing since world and all zones are spawned as a separate process. :)

Indeed, linux is pretty broad. We use PowerEdge 2850's and 2950's. And our issues have only been on the 2850's using Redhat. I understand from researching the issue is that the PERC cards are flashed to solve the issues, and that in some cases this doesn't solve it. It doesn't seem to have anything to do with the flavor of Linux as the issues that I have seen have been on different distros. It seems like it is related to how the particular interface on the PERC card reacts with the linux driver, but it has been a LONG time since I've done any driver development on Linux so can't say for sure.

We have yet to solve our issues as it is only a problem for us on boot up and we rarely boot the machines. Unfortunately for us, the problem machines are in a remote data center and require manual intervention when booting. We are looking at putting an IP KVM in there but it hasn't happened yet.

Are you compiling your own drivers or using the precompiled binaries from Dell?

I'm not certain to be honest. Those machines have been around a while, before I started working here and are in the UK so I don't really know how they were set up. I do believe that we run off of a canned kernel, which is mostly because of support issues.