DDraceNetwork - developer

DDraceNetwork

Development / developer

Development discussion. Logged to https://ddnet.tw/irclogs/ Connected with DDNet's IRC channel, Matrix room and GitHub repositories — IRC: #ddnet on Quakenet | Matrix: #ddnet-developer:matrix.org GitHub: https://github.com/ddnet

Between 2019-11-19 00:00:00Z and 2019-11-20 00:00:00Z

[ddnet-maps:master] 1 new commit

04da313 M Yoko 4.1 - ddnet-maps

What about to make fucking moving freeze?

3

3

HyperThonkSpin

2

What about to implement moving tiles instead just like gores did

@Anime.pdf kog already has moving freeze

There is a lot more that I want to get out of the way before we start adding new gameplay features

thats kog (edited)

15:09

what about ddnet

15:09

servers

@Anime.pdf there is a lot more that should done before we start adding new features (edited)

Hmmm

17:37

For example?

17:43

@n000b i saw u was typing...

Well look at the pull requests and issues on github. Besides there are some other projects which are not shown there (edited)

17:49

Learath wanted to change something with the ddnetnmasterserver to prevent ddos for example and he was trying to get a good solution for implementing accounts if i am right.

not to forget the update to 0.7

True

I want to give an stupid idea

18:21

Change masterservers ips

monkaS

he's too dangerious to be kept alive

18:27

btw no matter if you change the ip's, ddosers will ddos that specific ip too

18:29

ban ips that will sent too muck packages

@Anime.pdf attacker forges the udp packets so that the source ip is fake or from another legit player who did nothing wrong

Hmmm

20:11

What about sell this servers and buy servers with ddos protection? (edited)

it doesnt work like that

20:12

theres nothing u can pay for that would possibly protect tw

https://aws.amazon.com/ru/shield/ddos-attack-protection/ eeeeeh

Что такое DDOS-атака и как защитить св...

Узнайте, что такое DDOS-атака, разберитесь в классификации атак, а также в том, как защитить от них свой сайт.

20:14

Cant find us site oop

oop

cyka

20:14

and no

20:14

u dont get it

20:14

because of the way teeworlds netcode is written, and the way its fundamentally structured

20:14

there is no possible way to just buy a server that allows an unmodified tw server to handle such large ddoseses without lag

20:14

it requires code fixes

Fix the code?

20:15

:DD

its a better idea than throwing money at things u dont understand..

:c

20:16

Ehh

20:17

In which file net code? :V

network_server.cpp, network.cpp, server.cpp

20:17

and system.c

Oh yes

20:17

Its cpp

i dont claim to have the solutions for ddos, im slowly fixing my own server by editing mostly these files, but im sure a smarter person could do a better job with it

20:18

i just know that paying more isnt really the solution

20:18

these servers can easily receive all the ddos data without any lag

20:18

it only lags when trying to reply to everything

Umm

sorry

but where are that files, in what folder (im looking on github)

src/engine/shared/ for the first three and src/base/ for system.c

Thx

er

20:21

src/engine/server for server.cpp

20:21

src/engine/shared for the first two***

Ok

gl

Thx :D

Well, technically you could throw the money at a developer to fix it up

learath pls fix my server

20:25

$$$$$$$$$$$$$$$$$$$

If only I had the time and motivation for it

onby pls host server 20$ / month

im already hosting

20:26

and wow that could buy us a 4x more expensive server

:v

20:27

not that 20$ would be expensive, but this is how much i would cash out with some irl support

i mean

20:27

my sv costs $5/mon

We could get a more expensive server, but is that really the solution?

i really dont think it is

no, but dont deny money

because if the server can receieve all that data without lag

It'd just require one of us to get a living wage and sacrifice like 2-3 meals outside

then it should be able to reply smoothly too

ye more money = more smoothness feelsamazingman

feelsamazingman

Well there are fundamental limitations

if there was too much incoming data to overwhelm the link then it wouldnt even be able to receive it all without lagging

the issue is not only server performance but also limitations on packets per second

e.g. the linux network stack itself can't really do much more then 50kpps per core without some extra work

i mean when i run a server and let it receive all 200000 packets/sec of incoming junk, and use f2 to disable the replies to connect packets and inforeq packets, the server instantly stops lagging and anyone whos already connected can play smoothly like always

20:29

turning on the replies makes it lag when theres more than idk 100000pps

200kpps on a 5$/mon server?

mhm

How many vcores you get?

but only recv

20:30

1 lololol

20:30

i wish more

20:30

my threading code may be kinda useless otherwise

That's either some next level tuning on the kernel or the hypervisor

Image attachment

20:31

it can receive all this without lag if i make it stop sending out SendControl for tken and replies to fstd/gie3

20:32

but since those things make the server hidden and stop people from connecting its not really a fix

20:32

just a possible lead towards a solution maybe

Oh I missed that you don't reply at all

yeah

Sorry, just woke up

the replies is the problem

20:33

lol

Yeah, you can drop traffic a bit better then handling it

replying ofc means that you just double the pps count (edited)

Also recvmmsg is supposed to be helpful with this kinda thing

ddnet changed recv to recvmms

20:33

yeah

20:33

could we use the equivalent for sending

I don't think it'd be very useful anyway

20:34

It doesn't show up in profiling

thats a good sign isnt it

20:34

or u mean, the sending

20:34

Image attachment

The sending

some profiling output from when the server was receiving and trying to reply to all

20:35

threadrun is trying to reply to tkens

20:35

eh

20:35

i should do a profiling session without the experimental threading shit

You should also profile with callgrind

whats that

Stack sampling while nice is not that accurate

ooooo

Callgrind accurately profiles by dynamically rewriting the code

that looks cool

20:36

wtf xd

Not very performant

20:36

But it is very accurate

20:37

And overhead per function is same so all the costs are inflated equally

20:37

(Except for very small costs that end up getting amplified)

thx ill try this

woooow this is cool

21:23

Image attachment

21:23

Exported 130 message(s)