Server Warning: server CPU requirements increased!

[Utopia-Phoenix]

^^topic

I used to estimate the CPU requirements of one KF2 server room to be 0.6Ghz on Intel IVB. Now 0.8 is required.

I ran 24 server rooms on my server which consists of:

R5 3400G at 4.18 Ghz (at a voltage of 1.500V and a power consumption of 160 Watt running P95 small FFTs)
MSI B450M mortar
GEIL DDR4 16G 3200Mhz *4
Samsung PM981 256GB NVME

It works GOOD on 24 but SUCKS (laggy game performance) at 36. I expect to expand capacity by just upgrading RAM to 64GB from 32GB but reality spoke otherwise.

It could be calculated that such a 3400G would offer 4.18*4*1.3=21.736 (hyperthreading count as 1.3x)
The estimated demand of 36 rooms at 0.6 is 21.6 which is just up to its maximum, but it should work, not fail. It failed thus proving that the demand ratio is incorrect and need adjustment.

Careful server hosters!

 

[®omano]

I don't really understand your calculation. Also the requirement could vary a lot depending on the map the server is running, the number of players, the mods installed, and so on.

Something not in the equation is that a process can not split work on multiple cores, when it is single threaded (what KF2 servers are, right? mostly mainly?) so probably your math is wrong (if "0.2Ghz" is available on one core, then a KF2 server will have 0.2Ghz available to run not 0.2 here + 0.1 here + 0.4 here and so on.. if you see what I mean, anyway this is not very good way to calculate I think).

Also, the boost clock is not a all cores boost as far as I know, unless your manually overclocked for a "full OC all cores".

Max boost for AMD Ryzen processors is the maximum frequency achievable by a single core on the processor running a bursty single-threaded workload. Max boost will vary based on several factors, including, but not limited to: thermal paste; system cooling; motherboard design and BIOS; the latest AMD chipset driver; and the latest OS updates. Learn More.

So your cores are fluctuating from 3.7Ghz to 4.2Ghz not all at same time, and also depending on heat they may not even boost to 4.2Ghz at all.

Another thing, first sentence you talk about Intel IVB (Ivy Bridge?) then you say you are using R5 3400G.. so that doesn't make sense to me.

But yeah maybe KF2 server requirement increased since Early Access, that would make sense with all the stuff added over time.

 

[Utopia-Phoenix]

I don't really understand your calculation. Also the requirement could vary a lot depending on the map the server is running, the number of players, the mods installed, and so on.

Something not in the equation is that a process can not split work on multiple cores, when it is single threaded (what KF2 servers are, right? mostly mainly?) so probably your math is wrong (if "0.2Ghz" is available on one core, then a KF2 server will have 0.2Ghz available to run not 0.2 here + 0.1 here + 0.4 here and so on.. if you see what I mean, anyway this is not very good way to calculate I think).

Also, the boost clock is not a all cores boost as far as I know, unless your manually overclocked for a "full OC all cores".

So your cores are fluctuating from 3.7Ghz to 4.2Ghz not all at same time, and also depending on heat they may not even boost to 4.2Ghz at all.

Another thing, first sentence you talk about Intel IVB (Ivy Bridge?) then you say you are using R5 3400G.. so that doesn't make sense to me.

But yeah maybe KF2 server requirement increased since Early Access, that would make sense with all the stuff added over time.

You are right. To be precisely it is that Each server with 6 men DID require performance of 1 core of 0.6Ghz Intel IVB performance. A dual core 0.3Ghz IVB processor will NOT handle even a single server. It cannot be further split.

But note the following appendix:

1 With one PC running multiple servers, the OS would ultilize the servers into multiple threads. If you are carrying more than a handful of servers OR if you are NOT using ULTRA-LOW frequency/wattage platform (ultrabooks, SFF embedded solutions,etc) it could be considered as virtually perfect balanced between threads, therefore unneeded to calculate workload distribution in a single thread. Just calculate the server's total computation rate (TCR, measured in Ghz IVB performance) and make sure it surpasses the demand, that's enough.

2 The calculation of total computation rate is determined as Ghz IVBp. However there are no major difference between the IPC of XNB/IVB/Haswell/Broadwell/Skylake/Kabylake/Coffeelake AND RYZEN of all gens(1, 1+, 2). R5 3400G is ryzen 1+ and should be slightly superior in IPC compared to IVB. As a conservative policy I treated it as EQUAL.

3 I manually overclocked thus CPU to make sure all cores run at MAX freq at demand. That's the horrific record of 4.18Ghz at 1.500V. Passed P95 small FFT torture test of 24 hours plus.
I would surely NOT drop in such an inferior mistake of knowledge lol.

Therefore:
R5 3400G TCR: 4.18*4*1.3=21.736 Ghz IVBp (hyperthreading count as 1.3x)
Estimated requirements of 36 servers at 0.6G each: 21.6 Ghz IVBp
Estimated requirements of 36 servers at 0.8G each: 28.8 Ghz IVBp
Estimated requirements of 24 servers at 0.8G each: 19.2 Ghz IVBp

 

[oldschool]

Just an ignorant question on my part as the language alone is above my head.
In the config I remember there being an option to specify server usage per core? Would that help or not and why?
Thanks

 

[®omano]

I think you're talking about the CPU affinity that you can set in startup command line? If so, it is used to specify which CPU the server will only use.

-PREFERREDPROCESSOR=X where X is the core number https://wiki.killingfloor2.com/inde...(Killing_Floor_2)#Command_Line_Launch_Options

 

[doctorrock]

Hi.

My server CPU load graphs increased since October 1st, which correspond to the last update.


So I can confirm something has changed.

 

[®omano]

@Yoshiro any dev input about that?

For sure I noticed every single update brought something bad for performance on the client, but I can assume that server also got similar treatment.

 

[Yoshiro]

I am not aware of any change that would have impacted server reqs.

 

[®omano]

I don't monitor CPU usage and I personally didn't notice anything on my machine (which is kinda too powerfull for KF 2) but I see many more servers then before having lags/freeze when match start (sometimes frozen for five second when the pre(start counter hits 0), during trader (sometimes really horrible you can't even throw dosh for 2 seconds, people teleport and run like crazy fast when it 'unlag/unfreeze', and so on), and at end game too.

 

[doctorrock]

Well, I did notice that on a server of mine. It clearly has bigger load average than before the update.
Even when the server is idle waiting for people, it spikes the CPU at regular time, which it did not do before.