Do we really need 50 SPS?

[Alonzo]

I run Combat Box, a North America hosted server. It's been quite popular recently. We run fairly complex maps in order to create interest and realism. For most of last year we have been running on an overclocked 8600K in my basement. It's running at 4.8ghz coupled with fast 3200mhz memory. Server performance is good (tick times around 10-15ms at most) but the network isn't great -- it's hosted at my home on consumer-grade fibre. The 150/150 connection is pretty good, but it's not as good as hosting in a datacentre.

 

Last week we tried to switch to a dedicated 9700K in a professional datacentre in New York. Unfortunately because the server runs only at 4.3ghz with 2133 RAM, we immediately saw "multiplayer server overload" messages because the tick time crept up above 20ms. I have full performance data if people are interested, but to summarise: we were getting down as low as 30 simulations per second, with full-time "red blob bar" on the performance graph, and full-time "multiplayer overload" spam. So we switched away from this server back to the home hosted server.

 

But now players are complaining. They say that the player experience on the machine in the datacentre (with much better networking) was better than they see with the home-hosted server. This is despite the game running at 'only' 30 simulations per second and spamming with overload messages.

 

So my question: do we really need 50 simulations per second? Could we tune it to run at (say) 25? Or just suppress the 'overload' messages and manage server performance ourselves?

 

I'm stuck in a very frustrating catch-22. If I want datacentre hosting I have to put up with a slower CPU and RAM, which IL2 doesn't like, or I have to remove mission elements which I also don't want to do. But players are saying they'd rather be on the datacentre machine because it was a better experience, and they didn't notice the apparent overload! Help!

[RedKestrel]

On 2/3/2020 at 9:11 AM, Alonzo said:

I run Combat Box, a North America hosted server. It's been quite popular recently. We run fairly complex maps in order to create interest and realism. For most of last year we have been running on an overclocked 8600K in my basement. It's running at 4.8ghz coupled with fast 3200mhz memory. Server performance is good (tick times around 10-15ms at most) but the network isn't great -- it's hosted at my home on consumer-grade fibre. The 150/150 connection is pretty good, but it's not as good as hosting in a datacentre.

 

Last week we tried to switch to a dedicated 9700K in a professional datacentre in New York. Unfortunately because the server runs only at 4.3ghz with 2133 RAM, we immediately saw "multiplayer server overload" messages because the tick time crept up above 20ms. I have full performance data if people are interested, but to summarise: we were getting down as low as 30 simulations per second, with full-time "red blob bar" on the performance graph, and full-time "multiplayer overload" spam. So we switched away from this server back to the home hosted server.

 

But now players are complaining. They say that the player experience on the machine in the datacentre (with much better networking) was better than they see with the home-hosted server. This is despite the game running at 'only' 30 simulations per second and spamming with overload messages.

 

So my question: do we really need 50 simulations per second? Could we tune it to run at (say) 25? Or just suppress the 'overload' messages and manage server performance ourselves?

 

I'm stuck in a very frustrating catch-22. If I want datacentre hosting I have to put up with a slower CPU and RAM, which IL2 doesn't like, or I have to remove mission elements which I also don't want to do. But players are saying they'd rather be on the datacentre machine because it was a better experience, and they didn't notice the apparent overload! Help!

What is the practical impact of having fewer simulations per second? Presumably this means that the server is passing information less frequently to the client side (players). So would this impact things like bullet impacts or induce a lot more rubber-banding or stutters?

FWIW I was not able to get on the server last week, not sure if the stability issues were resolved but I was not able to connect early in the week and then haven't had time to get on since. 

Edited February 4, 2020 by RedKestrel

[Alonzo]

4 hours ago, RedKestrel said:

What is the practical impact of having fewer simulations per second? Presumably this means that the server is passing information less frequently to the client side (players). So would this impact things like bullet impacts or induce a lot more rubber-banding or stutters?

 

Good question. My observation from running a test (all single machine, no network latency) is that at some point the SPS becomes too slow for the client/server protocol to work properly, and you start to see planes moving with glitches. I'm guessing the client-side prediction of where other planes will be starts to be less effective, or the server itself is under a lot of stress and things are starting to go wonky. But that was probably down in the ~35 SPS range.

 

I imagine a lot of things are tied up in the 50 SPS number, probably both in single and multiplayer, and the Ai is coded to fly the planes at that level of granularity and so on. So it's probably not trivial to change, I had just hit a bit of a brick wall with this hosted server and was looking for other solutions.

 

Haluter has helped me eke out a little extra CPU performance and we've upgraded to faster RAM, both of which I hope will improve things, and I've also optimized some of our maps, so we hope to be able to squeeze the Combat Box experience into a little less CPU than usual.

[FTC_DerSheriff]

I really dont know. The overall flying was so much smoother on that data center. I personally havent seen many overload messages tho.
 

Edited February 5, 2020 by DerSheriff

[JG1_Butzzell]

S!

 

As your SPS drops....... strange things happen.

 

Check your cockpit clock.  Under sever low SPS the clock moves backward. It also affects the game program tracking the actual time. A session set for 2 hours could go many minutes longer. Have seen 15 minutes beyond what was set in ME.

[WWSitttingDuck]

have always thought the tick delay was more important than sps

 

Quote

SPS
Simulations per second (the number of simulation processing cycles per second). The
minimum value is 50. If the value drops below 50, the simulation slow-down becomes
noticeable.
Note: If the value exceeds 50, the SPS display still shows 50.

 

Tick Delay

The time required for one processing cycle of the simulation. The maximum processing
value must be less than or equal to 20 milliseconds (20ms). If the value is greater than
20ms, the simulation slow-down becomes noticeable

 

 

[Alonzo]

Just to round out this topic.

 

50 SPS is probably reasonable for an online game. Twitch FPS shooters run at least 64 ticks per second, and the newer ones run 128 (of course they're not doing so much physics processing). I still think you could run IL2 at 40 SPS and players wouldn't notice much, if any difference.

 

We ended up keeping Combat Box on the datacentre server, improving the RAM speed to 2666, and further optimizing our maps. It has been nice to brute-force the server using the 8600K, but the network connection on the New York server is much much better and definitely worth it. We still see some complaints from the server, but even when there are 'overload' messages players say they're not really noticing any lag. More noticeable is true network lag from the players, with some players jumping around a bit.

 

I'd still welcome any performance improvements on the server, as that allows us to have more ambitious missions, but we're doing ok on the current hardware setup.

[LLv34_Temuri]

On 3/8/2020 at 2:57 AM, Alonzo said:

I'd still welcome any performance improvements on the server, as that allows us to have more ambitious missions, but we're doing ok on the current hardware setup.

This. We are currently more in a "quite small skirmishes" level, not "great battles". They did improve the netcode a bit some time ago, I'm sure they can improve the performance too. They just need to find a way to squeeze in the performance improvements.