12 replies to this topic

[Jun Watarase]

Aside from using the optimal settings in TCPoptimizer, use it to disable nagling and checksum offloading (which is set to on for optimal settings). Im not sure if its the cause, but ever since I did that, im noticing much better hit reg especially with lasers.

[HammerMaster]

Jun Watarase, on 14 July 2019 - 08:58 AM, said:

Aside from using the optimal settings in TCPoptimizer, use it to disable nagling and checksum offloading (which is set to on for optimal settings). Im not sure if its the cause, but ever since I did that, im noticing much better hit reg especially with lasers.

"nagling" ?
Do you mean the RSC enable/disable?

Edited by HammerMaster, 14 July 2019 - 06:59 PM.

[MrVaad]

I think it's Nagle's algorithm. It's an algorithm to reduce the number of TCP packets sent over the network.

It's enabled by default on every OS. Disabling it *can* reduce your network speed but it *can* improve your latency Normally, it's enabled/disabled by the game/application for its connections.

Checksum offloading, i would not disable this, it frees CPU ressources by letting the network chip compute the packet's checksums.

Edited by MrVaad, 14 July 2019 - 11:35 PM.

[Jun Watarase]

I came across an article saying that checksum offloading can cause packets to get dropped so tried turning it off. Im pretty sure most modern CPUs wont have any trouble computing packet checksums anyway.

Edited by Jun Watarase, 15 July 2019 - 06:29 AM.

[Nightbird]

Cool, unfortunately, most complaints of hitreg comes from inability to aim, got an .exe to fix that?

[HammerMaster]

Nightbird, on 15 July 2019 - 08:58 AM, said:

Cool, unfortunately, most complaints of hitreg comes from inability to aim, got an .exe to fix that?

[Nightbird]

Jun Watarase, on 14 July 2019 - 08:58 AM, said:

Aside from using the optimal settings in TCPoptimizer, use it to disable nagling and checksum offloading (which is set to on for optimal settings). Im not sure if its the cause, but ever since I did that, im noticing much better hit reg especially with lasers.

https://www.speedgui...ettings-for-474

Any reasons to not follow this guide exactly?

[LordNothing]

anyone know where hit detect is being done? server or client side?

makes sense to have the clients handle collision detection for things like their own mech and weapons (and the terrain of course), and send those hit detection reports to the server for validation. its significantly faster to double check known collisions than to test every object in the game then distribute those qualified cd reports to the clients. its probibly done pure server side because engine limitations/unreliable client performance/pgi being a mini-dictatorship.

i considered something like that for my on game but i never really work on it anymore.

Edited by LordNothing, 15 July 2019 - 12:50 PM.

[LordNothing]

Nightbird, on 15 July 2019 - 08:58 AM, said:

Cool, unfortunately, most complaints of hitreg comes from inability to aim, got an .exe to fix that?

i trained a neural net to do it for me.



most of the hit detections im concerned with are high count scatter shot projectiles that likely overload network traffic and get culled. sometimes its better to have iffy culls than a temporary slowdown.

though i still swear by the fact that chain fired ppcs do more damage than group fired ones. seems like a completely different issue from above as a ppc dosent really tax the hit detect as much as lbs/missiles/racs. one projectile every four seconds. i imagine having ppcs get culled if somewhere else someone in a missile boat just fired a hundred shot volley at the exact time that i fired my guns. a cull priority issue might be to blame, seems you would want to cull lower damage projectiles first. there are other ways to reduce projectile overload like turn a 100 missile volley of 1 damage missiles into a 50 volly of 2 damage missiles.

you could also do geometric representations of projectile clouds and do a geometric intersect operation with the hit boxes to partition the damage appropriately. that could lead to the ppc behavior i see. ive also seen statistical approaches where there would be a look up table for damage percentage vs range or other parameters.

Edited by LordNothing, 15 July 2019 - 01:18 PM.

[MrVaad]

Also, If i'm not mistaken, the game servers only use UDP (like the majority of online games).

TCP is probably for login, the store, the news.

[LordNothing]

MrVaad, on 15 July 2019 - 01:12 PM, said:

Also, If i'm not mistaken, the game servers only use UDP (like the majority of online games).

TCP is probably for login, the store, the news.

i certainly don't want changes to my mech inventory to be done over udp. game packets its fine, most players wouldn't even notice a skipped network update. i wouldnt be shocked if pgi used tcp, but that pretty much doubles ping times (the need to send an acknowledge packet back), and my ping times seem more in line with udp than tcp.

Edited by LordNothing, 15 July 2019 - 01:26 PM.

[yrrot]

Plenty of games use both. UDP gets used for non-critical updates, or has a robust implementation to handle failed packets. TCP gets used for anything that has to be delivered, and has to arrive in order. It doesn't double ping, just increases it proportionately to how bad the connection is as it re-sends dropped packets. Your latency to the server is the same as far as actual trip times (roughly) between UDP and TCP, so your host state rewind, etc will always just use the one-way offset that works for both types.

I would guess that checksum offloading would cause packets to get dropped earlier in the chain, so they never go to the processor if they failed to deliver correctly. That should be faster (by some tiny ms) than doing checksum with the processor. Unless the other chip is super slow or something.

Nagle's algorithm may not matter one way or the other. Disabling it can increase bandwidth used, while slightly reducing the latency because packets get sent regardless of size asap, rather than queuing up to make a full packet. If MWO is using a decent structure on their packets, it may not change drastically. (like if they are always sending full-ish packets) Nagle's mostly is for reducing the impact of tiny packets with much larger headers. Like, sending 50 one byte packets has 2000 bytes of overhead, that could be sent in a single packet with 40 bytes of overhead.

[Monkey Lover]

Nightbird, on 15 July 2019 - 08:58 AM, said:

Cool, unfortunately, most complaints of hitreg comes from inability to aim, got an .exe to fix that?

20$ a month.