Dedicated Server - Resources

Hi everyone...


I'm about to switch root servers. Since I don't want to set up everything to often (server downtime is not appreciated by my friends) I want get a "proper" server right away.

I would like to run a PA server on it as well as

• Email
• Apache
• Samba
• TS
• Mumble
• Seafile
• VPN

I was looking at this offer: http://www.ovh.co.uk/dedicated_servers/kimsufi_16g.xml

Technology: Sandy Bridge
CPU: Intel Core i5 (4 cores / 4 Threads)
Frequency: 2.8GHz
RAM: 16 GB
Hard disk: 2TB SATA2
Bandwidth: 100Mbps
Location: France
OS: Debian or Ubuntu

What do you think, will this be sufficient?


Greetings
Slagathor aka Nutbag1911

It will be enough. But how big of a game do you want to run? Many gamers have a similar setup except the internet speed. I think you will be able to do, say up to 5000 units in game without an issue at release. (I heard they are shooting for 15k units per core, but I am estimating quite low) (bandwidth of 100Mbps should be enough for more than 5000 units)

sounds sufficient , thanks

What is uber running PA on at the moment?

They run game servers on cloud instances. Most of them in SoftLayer (I say it from my personal statistics), but probably not only. E.g AU server isn't in softlayer, but in other hosting provider.

We still know very very little about resource usage, when alpha started neutrino leave comment that game servers use one CPU core only, nothing else was said on that topic.

PS: There also was such comment from Uber that they're use Amazon EC2 too, but probably it's only was when alpha started. Actually you can easily check game server IP in logs and then check datacenter by whois.

I'd think it would depend heavily on how many sites you'll be hosting on it, how many email accounts, how much traffic you'll get on both, and potentially, how well optimized the websites are (in the event they get a decent amount of traffic).

If you're hosting 4 tiny websites then you basically have the full resources of the machine at PA's disposal, in which case it should probably be pretty good. If you're running 100 websites that get a reasonable amount of traffic and you're dedicating a big chunk of ram to MySQL then I don't know; I'd maybe be skeptical about the CPU in this case.

I take it for granted that it will become multithreaded at some point before release .

15k Units per core is a bit low tbh, the typical quadcore will pass out at 60k. Less if it has to run a bunch of other things too.
A 12 core will result in around 150-180k units depending on how well it scales, the more commonly found cheap gameservers wont be happy with 100k+. The million unit plus battles will need at least 16 cores and some further optimizations I guess?

How much bytes does the chronocam store per tick and unit? In huge 40 player battles that could well go into TB territory.

100MBit has to be good for 50k and more units if the 40 player mega games ever are to become real.

Where are you pulling these numbers from without knowing anything about the current hardware running PA, and without knowing about future optimizations?

I'm totally sure it's already multithreaded, probably from start of alpha because sim architecture it's not something you want completely change in alpha/beta.

They probably use only one core per server because it's let them run multiple game servers per one cloud instance without headache about CPU allocation. Don't forget they run servers on Windows currently, obviously for easier debug.

Basing on the blunt assumption of 15k per core as posted, if you have differing/better source please share some knowledge.
I have to admit that 15k would be somewhat low though.

If 50k chocke out a 100M link, guess what 1Mio will need. Therefore I just expect a 100Mbit line to suit 50k with ease.
As a 50k unit army is most likely commanded by less players than a 10^6 one - the big army needs much more bandwith than just 20 times that.
I hope 100M is good for far beyond 50k if the huge battles there is and/or was talk about are gonna be real.

And the storage for chronocam, well, that was a question and Im hoping for an answer .

There HAVE to be optimizations going much beyond 15k per core or 50k per 100M, no doubt about that.

Thats exactly what I had hoped to hear .