Testing server performances

Results:

The following graphs use avg values per real time minute to create datapoints.

Shiny graph 1 for the game on the AWS server, click for interactive version:

PA Stats: http://pastats.com/chart?gameId=230001

Shiny graph 2 for the game on softlayer:
PA Stats: http://pastats.com/chart?gameId=230171


Notice both have a few players that did not report to PA Stats, so the actual unit count is even a little higher.

Links to posts with some more info on stats of the servers:

AWS: https://forums.uberent.com/threads/testing-server-performances.65348/page-2#post-1022200

Softlayer:
https://forums.uberent.com/threads/testing-server-performances.65348/page-3#post-1022282

Overall this makes me think that having 4 cores is nice, but late in the game they all seem to be overloaded by whatever the servers does besides sim, so the sim most likely is not alone on 1 core, hurting the sim speed.
Though the game on the softlayer game happened to have a lot more units as well.
Looking at how the amazon server never used more than 5-6 cores in this game I think a high clocked hexacore might be best.

=====

Hey,

so to determine what kind of server we should get as community server I'd say we should do a few tests.
So I say let's say some 20-30 player FFAs tomorrow(?) to test server performance.
Currently my idea here is to compare the sim speed of the following servers:
- An amazon 16 core vm
- A quad Core Xeon 1270 V3: https://www.softlayer.com/Store/orderHourlyBareMetalInstance/37278/66
- And maybe the 24 core opteron

All these are either already online all the time or we can rent them for ~1$ an hour.

So for this we need 20-30 players interested into played a bunch of FFA. Preferably they should be set up in very similar ways and everyone should decide on one play style and play that one style in all test games, as to create comparable results.

The question is: Do we need 8+ cores for high player numbers or are 4 fast cores better?

So I would suggest tomorrow afternoon till noon for this. That means from the moment I post this thread in 20 hours, so 16:00 german time. Can we get enough people for this?

@Quitch also wanted to share this news

Meet at this time in the PA IRC
Have PA Stats installed please.

If I am around at the time (i'll be in IRC as usual), I'll happily play. But I can't guarantee it, stuff invariably happens I find!

What kind of connection do a 20-30 player game require? "server"

That is also a question we can answer with this. The amazon server has more than enough is all I know right now.
My guess is 1-4 mbit/s per player for upload.
Looking at your signature I wonder how many players fit in 36 mbit/s

hehe yeah , not many But I have been playing with the thought of running it as a server, with a little more OC. But the connection is not good enough for that, at this point.

I'm in

Spreading the word

Colin, you need to update the OP with a bit more information on what people require (PA Stats) and where to meeting (IRC? Monitor this topic?)

Any chance of a steam group for this? You can easily spam out gametimes and invites.

I'm up for it.

Done

I'll be here

Looking forward to these results.

I have some important stuff happening tomorrow afternoon, so I may or may not be able to make it. :-/

I don't think I'll be able to make it (but I'll try!) I'm rooting for the Xeon 1270 V3, I think it will outperform having more cores on the Amazon VM even with 20-30 people. I am hoping for numbers and graphs. I know there's a few people in the community with an unhealthy obsession with numbers and graphs, that could potentially spectate the games and take notes on sim speed, unit count, etc.

I'd love to join since a 30 person FFA is exactly my sort of organized chaos, but I kinda doubt I'll be able to handle the system, since my machine craps out and PA crashes trying to load the 'inner sol system' map. Is it gonna be generated or planned?

Does the ingame time counter actually slow down when the sim speed is low? If yes we can compare the final time it shows for the game with the time PA Stats shows. Though we need to make sure to subtract the landing phase, as PA Stats only starts to count once the first player gets his first economy "package".
Also does somebody know an easy to use tool to get some graphs of cpu memory and network usage on a headless ubuntu machine?

I am not even sure of that. A rather big game on pastats.com broke because it went out of memory, it didn't show any network lag. This is the result of a speed test on that server:

Not that much upload.

Confirmed

My PC is ready.

Even if not familiar with mods, I have installed PA Stats Online with PA Mod Manager.
I will try to be there for the big FFA.