Videos of current lag/consistency problems with the servers (rev. 3 by satori on 03-15-17, 03:08 pm)
Posted at 03-07-17, 04:44 pm Link | #1
satori

Posts: 92
Joined: 02-27-17
Last post: 1040 days
Last view: 453 days
edit: issue seems to be resolved, at least on 500 mtu
post rev. 1 by Flipend0 on 03-07-17, 05:21 pm
Posted at 03-07-17, 05:12 pm Link | #2
Flipend0 GM

Posts: 987
Joined: 02-06-17
Last post: 10 days
Last view: 10 days
It appears that you are using an AutoClicker for this
Please do not test latency with AutoClicker tools, as this messes up packet lining
post rev. 1 by satori on 03-07-17, 06:28 pm
Posted at 03-07-17, 06:14 pm Link | #3
satori

Posts: 92
Joined: 02-27-17
Last post: 1040 days
Last view: 453 days
It appears that you are using an AutoClicker for this
Please do not test latency with AutoClicker tools, as this messes up packet lining

both are with autoclickers and it still happens without using an autoclicker

there's something wrong but i'm not the one thats able to figure out if its the server or the routing or the game
post rev. 4 by Ohrami on 03-07-17, 06:21 pm
Posted at 03-07-17, 06:15 pm Link | #4
Ohrami

Posts: 20
Joined: 03-02-17
Last post: 2572 days
Last view: 2374 days
It appears that you are using an AutoClicker for this
Please do not test latency with AutoClicker tools, as this messes up packet lining
This is pretty simply refuted by two facts:
1. I experience similar lag despite having ~35 ms ping and not using an auto-clicker; not as good as 18 ms but it should still be much faster than what I got on the live servers with 100+ ping. Currently, it's noticeably slower.

2. Back in the live servers, a player with ~5 ping used an auto-clicker to prof wands and managed to do 5 heals in under a quarter of a second, perhaps faster; 5 heals looked exactly like 1 heal to the naked eye. Clearly, at least in the original version of Mabinogi, clicking extremely fast did nothing to increase latency.
post rev. 2 by Drahan on 03-07-17, 09:48 pm
Posted at 03-07-17, 09:34 pm Link | #5
Drahan GM

Posts: 2147
Joined: 02-06-17
Last post: 42 days
Last view: 7 days
I'd like to point out that there is a 10ms ping overhead between MabiPro's public IP address and our actual game server, so do take that into account when measuring ping from command prompt.

Local 0ms ping test server with Nagle's algorithm enabled
http://mabinogi.ir/img/vid/mabinogi_2017_03_07_001.webm

Local 0ms ping test server with Nagle's algorithm disabled (Theoretical fastest speed)
http://mabinogi.ir/img/vid/mabinogi_2017_03_07_002.webm

Remote 40ms ping production server with Nagle's algorithm disabled
http://mabinogi.ir/img/vid/mabinogi_2017_03_07_005.webm

There is a minuscule difference between production and test server as you can see by my videos and Nagle's Algorithm is a serious ping detriment so make sure that is turned off.

I don't see anything wrong with your 1500 MTU test, it is damn near close to the theoretical fastest speed.
The reason your skill actions are delayed is because of Mabinogi's built in packet queue.

Your 128 MTU test has inconsistencies that are caused by your local computer not dispatching packets as a whole. Basically, your MTU is too low. It's splitting up your packets into smaller segments which causes a significant delay during full transmission. MTU is not a packet queue size, lowering it does not cause packets to dispatch quicker.

There could be minor networking inconsistencies between you and the server, but that is completely out of our control and you should contact your ISP about that; I don't see any inconsistencies however.

Here's the point: As long as your client is sending packets properly, packets are arriving in a timely manner, and you don't hit Mabinogi's built in packet queue that delays packets intentionally if you send too much (like autoclicking), you will not experience lag.
post rev. 1 by Ohrami on 03-07-17, 10:44 pm
Posted at 03-07-17, 10:42 pm Link | #6
Ohrami

Posts: 20
Joined: 03-02-17
Last post: 2572 days
Last view: 2374 days
Why, though, is the "theoretical maximum speed" on this server so much slower than it was on the live servers back in the day? Like I said, people with 5 ping were able to literally heal 5 times so fast that it looked like a single heal. Another thing I'd like you to test is casting icebolt (or any bolt) onto a monster while you are not in combat mode. Initializing combat mode seems to be one of the laggiest functions in the game, but anyone who played the game back then can distinctly remember all the Californians casting bolts immediately after they finished charging them even when they weren't starting in combat mode. You can even see in the 100+ ping video in the OP that her speed is comparable to your "theoretical maximum" video, albeit less consistent.

There definitely seems to be some sort of an issue with how the server processes data compared to how it was in the original game.
post rev. 1 by Drahan on 03-07-17, 10:44 pm
Posted at 03-07-17, 10:44 pm Link | #7
Drahan GM

Posts: 2147
Joined: 02-06-17
Last post: 42 days
Last view: 7 days
Why, though, is the "theoretical maximum speed" on this server so much slower than it was on the live servers back in the day? Like I said, people with 5 ping were able to literally heal 5 times so fast that it looked like a single heal. Another thing I'd like you to test is casting icebolt (or any bolt) onto a monster while you are not in combat mode. Initializing combat mode seems to be one of the laggiest functions in the game, but anyone who played the game back then can distinctly remember all the Californians casting bolts immediately after they finished charging them even when they weren't starting in combat mode.

There definitely seems to be some sort of an issue with how the server processes data compared to how it was in the original game.

I can test that later, but I'd like to point out we are the "original game", considering we run the server with the real files and we're not an emulator.
Posted at 03-07-17, 10:46 pm Link | #8
Ohrami

Posts: 20
Joined: 03-02-17
Last post: 2572 days
Last view: 2374 days
Why, though, is the "theoretical maximum speed" on this server so much slower than it was on the live servers back in the day? Like I said, people with 5 ping were able to literally heal 5 times so fast that it looked like a single heal. Another thing I'd like you to test is casting icebolt (or any bolt) onto a monster while you are not in combat mode. Initializing combat mode seems to be one of the laggiest functions in the game, but anyone who played the game back then can distinctly remember all the Californians casting bolts immediately after they finished charging them even when they weren't starting in combat mode.

There definitely seems to be some sort of an issue with how the server processes data compared to how it was in the original game.

I can test that later, but I'd like to point out we are the "original game", considering we run the server with the real files and we're not an emulator.
Right, I understand that it's using the original files, but I imagine that there may be some sort of server setting or something not configured correctly which is leading to this inconsistency/lag. There is no question whatsoever that the theoretical maximum speed you've proposed in your videos is much slower than the speeds tons of people experienced daily back in G8-G13.
Posted at 03-07-17, 10:47 pm Link | #9
Drahan GM

Posts: 2147
Joined: 02-06-17
Last post: 42 days
Last view: 7 days
Why, though, is the "theoretical maximum speed" on this server so much slower than it was on the live servers back in the day? Like I said, people with 5 ping were able to literally heal 5 times so fast that it looked like a single heal. Another thing I'd like you to test is casting icebolt (or any bolt) onto a monster while you are not in combat mode. Initializing combat mode seems to be one of the laggiest functions in the game, but anyone who played the game back then can distinctly remember all the Californians casting bolts immediately after they finished charging them even when they weren't starting in combat mode.

There definitely seems to be some sort of an issue with how the server processes data compared to how it was in the original game.

I can test that later, but I'd like to point out we are the "original game", considering we run the server with the real files and we're not an emulator.
Right, I understand that it's using the original files, but I imagine that there may be some sort of server setting or something not configured correctly which is leading to this inconsistency/lag. There is no question whatsoever that the theoretical maximum speed you've proposed in your videos is much slower than the speeds tons of people experienced daily back in G8-G13.

There is no server settings related to networking, in fact most of the server isn't really configurable by normal means.
post rev. 1 by Ohrami on 03-07-17, 10:54 pm
Posted at 03-07-17, 10:51 pm Link | #10
Ohrami

Posts: 20
Joined: 03-02-17
Last post: 2572 days
Last view: 2374 days
I don't know how the game is now, but back in G13 and earlier, the actually achieved speeds were significantly higher than your proposed theoretical maximums. This means that you either performed the test incorrectly or that there is some sort of difference between the files which were actually used for the original game and what you are using currently. You should also test to see if elf charge+smash is possible without the opponent breaking out of hit-stun, as this is a much more visible test and its possibility is very well-documented.
Posted at 03-07-17, 10:55 pm Link | #11
Drahan GM

Posts: 2147
Joined: 02-06-17
Last post: 42 days
Last view: 7 days
I don't know how the game is now, but back in G13 and earlier, the actually achieved speeds were significantly higher than your proposed theoretical maximums. This means that you either performed the test incorrectly or that there is some sort of difference between the files which were actually used for the original game and what you are using currently.

The files we are currently using were used on original Mabinogi during G13S1-G13S2 for Japan test server.

Tested icebolt, there is no noticable difference between 0ms and 40ms.
The lockup seems to be clientsided, so it was probably a mod that fixed it.

Here they are:
0ms: http://mabinogi.ir/img/vid/mabinogi_2017_03_07_011.webm
40ms: http://mabinogi.ir/img/vid/mabinogi_2017_03_07_009.webm
Posted at 03-07-17, 10:58 pm Link | #12
Ohrami

Posts: 20
Joined: 03-02-17
Last post: 2572 days
Last view: 2374 days
Could you also test to see if elf charge+smash is even possible? I don't play an elf and I'm busy with real life stuff and will be for the next couple weeks otherwise I would test a lot of this myself.
Posted at 03-07-17, 11:00 pm Link | #13
Drahan GM

Posts: 2147
Joined: 02-06-17
Last post: 42 days
Last view: 7 days
Could you also test to see if elf charge+smash is even possible? I don't play an elf and I'm busy with real life stuff and will be for the next couple weeks otherwise I would test a lot of this myself.

Is elf Charge somehow different than normal Charge?
post rev. 7 by Ohrami on 03-07-17, 11:11 pm
Posted at 03-07-17, 11:01 pm Link | #14
Ohrami

Posts: 20
Joined: 03-02-17
Last post: 2572 days
Last view: 2374 days
I have another source showing from the official Mabinogi channel that someone who presumably has a very low latency (I imagine inside of Nexon HQ) was able to nearly instantly cast icebolt from non-combat mode with next to no delay: https://www.youtube.com/watch?v=f7RGO3ChI90

I'm assuming that the official Mabinogi channel wouldn't be using mods in their video. I also assume that they likely didn't have Nagle's algorithm disabled, which explains the very short (but still noticeable) delay. Some players from California would have absolutely no noticeable delay whatsoever on this type of attack. Perhaps the Japanese and NA clients are different, with the NA client more optimized for online play, but I couldn't imagine why.

Edit: Actually, I've seen several Japanese players in some old videos, and I distinctly remember them being very lagless. Will try to dig up some of these source videos.
Posted at 03-07-17, 11:02 pm Link | #15
Ohrami

Posts: 20
Joined: 03-02-17
Last post: 2572 days
Last view: 2374 days
Could you also test to see if elf charge+smash is even possible? I don't play an elf and I'm busy with real life stuff and will be for the next couple weeks otherwise I would test a lot of this myself.

Is elf Charge somehow different than normal Charge?
No, but their Smash attack is. Charge has a 2.6 second stun and elf Smash takes 2.5 seconds to charge. The real test for a lagless player (especially in PvP) was to be able to consistently use Charge and follow it up with Smash as an elf without your opponent breaking out of hit-stun. It's pretty well-documented and every person who did a lot of dueling in the main game knows about it.
Posted at 03-07-17, 11:23 pm Link | #16
satori

Posts: 92
Joined: 02-27-17
Last post: 1040 days
Last view: 453 days
all of my videos are with leatrix latency fix enabled (disables nagle's algorithm), the very first video is from G13 or G14 on the live nexon server

the inconsistencies exist and are more obvious on an offline server:
https://youtu.be/cm9EXXf7ZR8

there are pauses while spamming lb, healing, weapon switching and resting
i assure you these didn't exist in the live server

everything is much slower compared to socal/texan players from the live G13, in terms of weapon switching, skill spamming, rest spamming, etc

i won't post on this topic anymore i just wanted to inform people
post rev. 1 by Drahan on 03-07-17, 11:29 pm
Posted at 03-07-17, 11:24 pm Link | #17
Drahan GM

Posts: 2147
Joined: 02-06-17
Last post: 42 days
Last view: 7 days
all of my videos are with leatrix latency fix enabled (disables nagle's algorithm), the very first video is from G13 or G14 on the live nexon server

the inconsistencies exist and are more obvious on an offline server:
https://youtu.be/cm9EXXf7ZR8

there are pauses while spamming lb, healing, weapon switching and resting
i assure you these didn't exist in the live server

everything is much slower compared to socal/texan players from the live G13, in terms of weapon switching, skill spamming, rest spamming, etc

i won't post on this topic anymore i just wanted to inform people

That inconsistency is the mint/packet queue delay, which is clientside, not server lag.
post rev. 1 by Drahan on 03-07-17, 11:30 pm
Posted at 03-07-17, 11:28 pm Link | #18
Drahan GM

Posts: 2147
Joined: 02-06-17
Last post: 42 days
Last view: 7 days
Tested elf Charge + Smash;

it is easy on 0ms ping and mildly harder on 40ms with about 10-15ms (nearing the stun completion animation) left in the mob's stun before the Smash finishes charging.

0ms: http://mabinogi.ir/img/vid/mabinogi_2017_03_07_014.webm
40ms: http://mabinogi.ir/img/vid/mabinogi_2017_03_07_020.webm

These tests and what Satori has shown just proves that it's absolutely not a networking issue or a testing issue.
post rev. 2 by Ohrami on 03-07-17, 11:36 pm
Posted at 03-07-17, 11:32 pm Link | #19
Ohrami

Posts: 20
Joined: 03-02-17
Last post: 2572 days
Last view: 2374 days
I found a decent source for how 10 ms should look like. Keep in mind this is a 2016 test, so the client obviously would have changed since now, but I can pretty much guarantee people around this lagless played in 2010. https://www.youtube.com/watch?v=e7XXtrMu-6s

This is undeniably less laggy than what you showed even at 0 ping.

Edit: I'm wondering if perhaps the mods/hacks he's mentioning in the start of the video are drastic enough differences to make up for him having ~10 ms of network lag. I'm going to look into it a little bit
post rev. 2 by Drahan on 03-07-17, 11:38 pm
Posted at 03-07-17, 11:36 pm Link | #20
Drahan GM

Posts: 2147
Joined: 02-06-17
Last post: 42 days
Last view: 7 days
I found a decent source for how 10 ms should look like. Keep in mind this is a 2016 test, so the client obviously would have changed since now, but I can pretty much guarantee people around this lagless played in 2010. https://www.youtube.com/watch?v=e7XXtrMu-6s

This is undeniably less laggy than what you showed even at 0 ping.

That's completely impossible to do without modding the game. He has modified the networking code of the Mabinogi client; he removed the clientside mint/packet queue that I keep talking about, and optimized a lot of it to get it to function that way.

That is absolutely not what 10ms *should* look like at all.
Terms

Powered by mabi.pro v1.0034-arisa (View credits)
MabiPro is not associated with Nexon Co., Ltd. in any way shape or form.