Tree of Savior Forum

I checked my GPU load during terrible city framerate loss. Here's what I discovered

That’s 0.1 ~ 0.2 SECONDS.

Just for reference.

*fixed my post
yeah of course… ms would be very impressive for that!

It’s 5 am here, so bear with me :wink:

Testing progress:

  • Chat spam lags, regardless of chat bubbles on or the chat window being displayed. Length of messages directly affect how hard the fps drops are.

  • Players degrade performance. I was getting 45fps by the TP Trader (with 6 other onscreen players); moving southwest to where there isn’t a single NPC (and one other player) instantly put me back to 60fps.

  • @chronosanct spammed some Dievdirbys skills (the ones that have some field effects), overlapping a lot of them didn’t drop my fps even a bit.

  • Spamming poses doesn’t affect FPS either.

1 Like

@Staff_Julie

Yeah I would suggest at least a half second chat timeout period (for even normal chat). Because you can spam long strings HARD and really tank FPS of other people nearby currently…

I can confirm this. It does happens with me. If I stay too long in a map, like grinding, the weapon switching for Concentrated Shot tooks too long and sometimes I can’t even see anything of the skill until it finishes.

Also, dunno if it’s related. The FPS starts dropping even more if I have 30% or less durability on the items.

I get like 14 fps in a town… and in a non crowded map, obo 30 fps (i7 4702HQ, gtx 765m, 8gb DDR 3 ram).

Or a 140 byte limit on messages. If it’s fine for Twitter, it’s fine for your everyday in-game conversation.

That and I find half second too little; I’d say at least 2 seconds because that’s as fast as one will type a message, and those who use the Enter key as punctuation will have to learn to type correctly.

I’m not talking about graphical processing.

I’m talking about damage computation (skill damage + stat + eq + whatever - monster resist - monster defense - monster whatever)

It’s like throwing an ajax request and your application just sits in loading state.
It’s like on the server side, they make monster db queries, item db queries, skill db queries -> perform computations -> return data to client. Hence why a character just freezes it’s skill/attack animation -> sometimes i can cancel it like the request was a lost packet.

I’m a web programmer and I know a little bit about client-side(browser) vs serve-side(server) computation and how to optimize these.

And it depends mainly on a use case such as “skills” or “attacks”, etc. Whether to partially doing client-side computation, client side caching, etc.

You can cache and float data in memory so client makes less server processing. I opened a topic regarding this and “lock step” theories.

Graphical processing is OBVIOUSLY not server-sided; that would be impractical, and no one would have enough bandwidth to play the game.

Damage computation is really, REALLY light on the server, and something like a Core2Quad Q9550 should be able to handle it for tens of thousands of players.

And again, they use AWS. Their server platform is really flexible, and if IMC was having any server bottlenecks, they probably would already have provisioned better servers as the performance issues would be a lot more severe than they are right now, and having an overloaded CPU can cause some verification routines not to run, making item duping possible in some cases etc…

Did you read the part with the potential db queries they are making that’s causing these delays?

On top of that, I’ve experience rubberbanding.

Non of which I experience while playing Diablo3 on NA. My point is, there is room for optimization server-side, and not just client-side.

DB queries would be an issue for storage read speeds…

… if AWS didn’t have their whole storage off PCIe SSDs, right?

Very few of the issues we’re experiencing right now would be solved with server-side optimization, and the skill example you used is usually caused by actual packet loss (feel free to Wireshark the game and see if the TCP ACK packet ever arrives).

By the way, Blizzard is a company that’s considerably larger than IMC, with more money to invest in servers, and, guess what!? A smaller playerbase! (at least on your D3 example; WoW clearly has more players than ToS does)

Smaller? Regardless of when, here’s how to fact check.

TOS is at 22,401 (I checked on steam stats).

I honestly don’t know what you’re trying to argue… All I’m saying is that ON TOP OF the graphical client side optimizations that needs to be done, they also need to fix their server side computing.

But true, majority of the issues are client side. Second would be their business decisions. Third would be server side optimization.

All I’m saying is, there ARE server side issues. I’m not saying there are more compared to client side. I get your point - finally lol. But don’t make up statistics about playerbase. lol

That’s three years old data; when the game was at its peak. It’s pretty much dead right now (and it counts all regions! The game is the most popular at Asia, and we’re talking about the NA server, right?)

Besides, you’re comparing unique daily player metrics with simultaneous player metrics, which are largely different.

I still think it might be a memory leak, knowing that it has a long term effect.

The terrible performance has little to do with their servers, ■■■■ was the same on kToS when I played there as well. The game barely uses your gpu because it doesn’t have any complex lightning, 3d rendering or physics. It doesn’t need to.

Most performance issues in MMOs are CPU based or network based, but since even the top of the line single threaded performance CPUS still struggle to deliver any reasonable framerate in boss fights I think it’s pretty obvious the issue is with their netcode. I say this because Granado Espada was running on a similar engine and it was the same thing the moment you had multiple characters around (even worse since each player had 3 characters). I don’t know any specifics but the amount of traffic it gets sent is trough the roof and it’s easy to confirm with wireshark or something similar.

It’s also easy to realize how your client would start freezing back when goldsellers used to spam the mic. It’s clearly network related.

1 Like

I just said I agree with you. Still arguing… And now you’re off topic. Some people argue for the sake of arguing. I’m moving on from this. Enjoy your day.

PS That’s why I said regardless of date.

reading this keyword made me remember that usual mmo issue registry fix, regarding
TcpAckFrequency and TCPNoDelay.
I have both set on my system. So this should actually reduce lag in some regard.
But in RO it seems it only lagged monster positioning not let the whole game wait/freeze/slow down.

This has become an apparent issue with the skill Firewall on undead monsters in raganrok online.
here are two video examples of how it changed things for that game:

Windows Default settings:

TcpAckFrequency set to 1:

Judging by videos the one with TCPAckFrequency is better. Undead monsters in RO are not affected by firewall’s knockback effect so firewall is a great burst to kill them. The 1st video looks laggy and monster positioning gets bugged.

It seems to be a huge problem with caching as well (opening up map, mini-map lags the entire game). This might be because the disk-load is very high. One (can’t at the moment) should try loading the entire game with a RAMDISK to check how it affects performance.

I think i tried this during iCBT2, but back then it didn’t change all to much for me if i remember correctly.