Analysis of NA, EU, KR, Local latency from NA

bwally

United States670 Posts

July 12 2011 23:57 GMT

Description:
Ok so basically I have seen many questions regarding latency across different servers so I decided to create a video demonstrating the latency from Myself->NA, Myself->EU, Myself->KR, Single Player aka Local, and "cracked" LAN. I am located on the East Coast (NY) of the USA. In addition to the video I added some "empirical" evidence/numbers.

Not sure if anyone has done all of these tests but here it is anyway.

Details:
These are the blizzard SC2 logon servers for each region, they may be a different entity than actual game servers, but I figure they are probably close in distance. There might be others that will automatically adjust for your location but I have no evidence or way to find this at the moment.

NA: 12.129.206.130 (us.logon.battle.net) located in CA, USA
EU: 213.248.127.130 (eu.logon.battle.net) located in Paris, France or UK? (Close enough regardless)
KR: 121.254.200.130 (kr.logon.battle.net) located in Seoul, Korea

Since I can't directly send ICMP packets (Ping) to each IP I did a speedtest.net ping to servers located in the same country/state.

The results for me are this (round-trip time):
NA: 70-80ms
EU: 85-120ms
KR: 210-242ms (Ouch)

Traceroutes:
+ Show Spoiler [NA] +

21 79 ms 78 ms 80 ms mdf001c7613r0003-gig-10-1.lax1.attens.net [12.129.193.242]
22 12.129.211.34 reports: Destination net unreachable.

+ Show Spoiler [EU] +

8 285 ms 10 ms 18 ms nyk-bb2-link.telia.net [213.155.130.244]
9 110 ms 109 ms 112 ms ffm-bb2-link.telia.net [213.155.131.150]
10 116 ms 118 ms 117 ms ffm-b10-link.telia.net [80.91.251.124]
11 * * * Request timed out.
12 * * * Request timed out.
13 * * * Request timed out.
14 * * * Request timed out.
15 * * * Request timed out.
16 * * * Request timed out.
17 * * * Request timed out.
18 * * * Request timed out.
19 * * * Request timed out.
20 * * * Request timed out.
21 * * * Request timed out.
22 * * * Request timed out.
23 * * * Request timed out.
24 * * * Request timed out.
25 * * * Request timed out.
26 * * * Request timed out.
27 * * * Request timed out.
28 * * * Request timed out.
29 * * * Request timed out.
30 * * * Request timed out.

+ Show Spoiler [KR] +

12 83 ms 80 ms 79 ms te0-0-0-5.ccr21.lax01.atlas.cogentco.com [154.54.0.221]
13 80 ms 88 ms 79 ms te9-1.ccr02.lax05.atlas.cogentco.com [154.54.44.130]
14 79 ms 81 ms 80 ms 38.104.84.42
15 190 ms 189 ms 189 ms gi12-0-0.cr2.nrt1.asianetcom.net [202.147.0.125]
16 225 ms 223 ms 226 ms po3-0.gw4.sel2.asianetcom.net [202.147.8.73]
17 210 ms 211 ms 210 ms LGT-0001.gw3.sel2.asianetcom.net [202.147.9.18]
18 232 ms 233 ms 243 ms 1.208.104.245
19 213 ms 213 ms 212 ms 211.233.55.126
20 212 ms 212 ms 210 ms 211.234.120.150
21 121.254.199.42 reports: Destination net unreachable.

Video:
I created this video showing each server "test", it's quite bland :D

Timestamps:
NA - 00:00 to 02:08
EU - 02:08 to 04:30
KR - 04:30 to 07:54
Local - 07:54 to 10:32

Analysis:
So after I created the video I went back and did a frame by frame analysis to determine a rough latency. I used the rally point for most samples since it seemed to be the easiest to tell when the "reaction" happened. Also I am pretty sure there is no acceleration property of movement like a unit would have, slow ass Overseer, therefore setting the rally point should be as fast as possible.

So basically the first sign of the 4 greens arrows is where I set the "start" point frame. Shown Below
[image loading]

and I stopped on the first frame where the rally point changed. Shown Below
[image loading]

+ Show Spoiler [Latency Samples] +

NA:
Click @ frame 2686
Reaction @ frame 2694
Delay = 8 frames

Click @ frame 2794
Reaction @ frame 2800
Delay = 6 frames

Click @ frame 2830
Reaction @ frame 2837
Delay = 7 frames

Click @ frame 2886
Reaction @ frame 2893
Delay = 7 frames

EU:
Click @ frame 5210
Reaction @ frame 5219
Delay = 9 frames

Click @ frame 5269
Reaction @ frame 5278
Delay = 9 frames

Click @ frame 5325
Reaction @ frame 5334
Delay = 9 frames

Click @ frame 5387
Reaction @ frame 5394
Delay = 9 frames

KR:
Click @ frame 11599
Reaction @ frame 11614
Delay = 15 frames

Click @ frame 11839
Reaction @ frame 11854
Delay = 15 frames

Click @ frame 11978
Reaction @ frame 11991
Delay = 13 frames

Click @ frame 12533
Reaction @ frame 12546
Delay = 13 frames

Local:
Click @ frame 15648
Reaction @ frame 15649
Delay = 1 frame

Click @ frame 15680
Reaction @ frame 15682
Delay = 2 frames

Click @ frame 15735
Reaction @ frame 15736
Delay = 1 frame

Click @ frame 15983
Reaction @ frame 15984
Delay = 1 frame

Cracked LAN (Separate Video Source):
Click @ frame 375
Reaction @ frame 382
Delay = 7 frame

Click @ frame 403
Reaction @ frame 411
Delay = 7 frame

Click @ frame 433
Reaction @ frame 442
Delay = 9 frame

Click @ frame 468
Reaction @ frame 475
Delay = 7 frame

So here are the average frame delay:

NA: 7 frames
EU: 9 frames
KR: 14 frames
Local: ~1.25 frames (really don't have this precision but whatever)
Cracked LAN: ~7.5 frames

The video is 29.970 fps (frames per second) which means:
1 frame = (1sec / 29.970fps) * (1000ms / 1sec) = 33.37 ms (milliseconds), per frame

Conclusion:
For my location (USEast) to these servers, can be looked at as round-trip time:

NA: 233.59 ms avg
EU: 300.33 ms avg
KR: 467.18 ms avg
Local: 41.71 ms avg
Cracked LAN: 250.27 ms avg

Regarding "Cracked" LAN: My guess is that the code/server routine is not optimized at all. It does have to emulate the async aspect of the engine and probably does it as slow as possible to not throw desync type errors (no idea though, haven't analyzed the emu server enough). I have had to do the test with 2 SC2 clients open but it didn't seem to slow anything. This also doesn't mean that blizzard's implementation of LAN would be relativity "slow" as well or a modified version of this emu could have a lower latency or there is a hard minimum latency that cannot be reduced, however this is just speculation.

NA vs Local: 82.1% less latency than on NA
NA vs EU: 28.5% more latency than on NA
NA vs KR: 100% more latency than on NA

EU vs KR: 55% more latency than on EU

To explain the percentages let's say if you have 40ms vs 80ms on 2 different regions you probably couldn't tell the difference in a game but 80ms is actually 100% more laggy than 40ms. So you have to know at which point this time becomes noticeable.

Since I record all the samples in the same way looking at the actual delay number is probably not as useful as looking at the delta time between each region. Of course this might not be true if you're talking about really high latencies like 400ms+ which would feel much worse. Then you can say that this region is XX% times worse than your normal region.

Now there might be some errors created from recording at a low framerate and you could get an even more accurate number by increasing the recording framerate but I think these numbers are good enough.

Reading:
If you want to know why ICMP ping is not really as useful as this measurement read this.

dartoo

India2889 Posts

July 13 2011 00:02 GMT

You could do a tracert to each blizzard server right? That might be more cleaner way of doing it..

Edit:Hmm..why couldnt you send packets to the ip? i dont understand..if you couldnt send packets then how can you play?

bwally

United States670 Posts

July 13 2011 00:06 GMT

On July 13 2011 09:02 dartoo wrote:
You could do a tracert to each blizzard server right? That might be more cleaner way of doing it..

Edit:Hmm..why couldnt you send packets to the ip? i dont understand..if you couldnt send packets then how can you play?

The actual ping doesn't matter too much because there is inherent latency from how the game engine works and the latency I "calculated" is more meaningful. But you would get something like this trying to ping, Blizzard blocking ping requests I guess:

ping 12.129.206.130
Pinging 12.129.206.130 with 32 bytes of data:
Reply from 12.129.211.34: Destination net unreachable.
Reply from 12.129.211.34: Destination net unreachable.

The latency I "calculated" is based on seeing actual game events so I think it's pretty good.

Mohdoo

United States15690 Posts

July 13 2011 00:12 GMT

This is a wayyyy better way of testing lag on each server, mad props, OP!

lazydino

Canada331 Posts

July 13 2011 00:47 GMT

My NA-NA latency is equivalent to your NA-KR latency. Fucking telus

alexhard

Sweden317 Posts

July 13 2011 01:06 GMT

242ms only appears to be much worse than the others. The fact is that anything below 200ms is irrelevant, because that's the minimum that the hosting servers enforce. 42ms above that is sub-optimal of course, but not really that bad.

OTOH, why is your local latency that high? Is it an artifact of the measurement method or is the game simply that clunky?

Lunares

United States909 Posts

July 13 2011 01:09 GMT

On July 13 2011 10:06 alexhard wrote:
242ms only appears to be much worse than the others. The fact is that anything below 200ms is irrelevant, because that's the minimum that the hosting servers enforce. 42ms above that is sub-optimal of course, but not really that bad.

OTOH, why is your local latency that high? Is it an artifact of the measurement method or is the game simply that clunky?

Except if you look at measured latency not just ping latency korea is at almost 400ms, which is a significant amount for sure.

bwally

United States670 Posts

July 13 2011 02:02 GMT

On July 13 2011 10:09 Lunares wrote:

Show nested quote +

Except if you look at measured latency not just ping latency korea is at almost 400ms, which is a significant amount for sure.

Yea, exactly like I said this is a more meaningful number than ping. Because this latency is everything the game engine adds in addition to the physical limitation of protocol specs, network hardware, network routing, and speed of light.

Also the resolution I have is ~33ms due to 1 frame increments equaling 33ms at the recorded framerate. So the local latency could possibly be less if you simply record it at a high framerate to find the lowest latency possible.

Interesting article on why you have these inherent delays: http://altdevblogaday.com/2011/07/09/synchronous-rts-engines-and-a-tale-of-desyncs/

mapthesoul

Trinidad/Tobago429 Posts

July 13 2011 02:13 GMT

Would love to see an analysis of EU to KR.
It's pretty laggy EU to US in my experience.

DirtYLOu

575 Posts

July 13 2011 02:16 GMT

#10

Way to go.

60ms on NA LOL

1ms lan latency plox.

Azzur

Australia6259 Posts

July 13 2011 02:16 GMT

#11

Excellent way of analysing latency! I've got to try this

Pulzlulz

Germany412 Posts

July 13 2011 02:40 GMT

#12

Would you consider trying out the cracked LAN version for a Single Player - LAN comparison?

ehalf

408 Posts

July 13 2011 03:33 GMT

#13

Interesting. I also live in the US east and my ping to the servers are:
NA:80
EU:150
KR:350

Play on ping with 150 is sometimes noticeable but generally hardly affect the gameplay. But 300ms ping is killing me,

aksfjh

United States4853 Posts

July 13 2011 03:38 GMT

#14

On July 13 2011 11:16 DirtYLOu wrote:
Way to go.

60ms on NA LOL

1ms lan latency plox.

I don't think you can even get that with somebody down the street unless you dumped almost all protocols and parity out the window.

Skyze

Canada2324 Posts

July 13 2011 04:52 GMT

#15

So far in the 5 games I've played on TW server, the lag seems alright. Better than me playing on EU honestly. Theres afew spikes, but mainly its pretty consistent, and a lot less hiccups than EU. Granted, this is only 5 games as opposed to ~150 ive played on EU, so we'll see when I play more

(im in Eastern Canada)

OmniscientSC2

United States713 Posts

July 13 2011 05:12 GMT

#16

Dope. I really wish I could get LAN Latency in ZvZs though =\ Makes Ling/Baneling so much easier with manual detonation. I've noticed that my lag isn't too bad from US to KOR all the time. It's decent until peak hours which is when it gets a bit unplayable.

Benga

Korea (South)471 Posts

July 13 2011 08:17 GMT

#17

Lol how could the koreans compete and mostly win the tournaments in Nasl,TL Open...

arbitrageur

Australia1202 Posts

July 13 2011 11:34 GMT

#18

So if the tracert to kr.logon.battle.net results in 400ms near the end is purchasing a KR account a waste of $?

brrrtmn

Russian Federation154 Posts

July 13 2011 12:15 GMT

#19

On July 13 2011 20:34 arbitrageur wrote:
So if the tracert to kr.logon.battle.net results in 400ms near the end is purchasing a KR account a waste of $?

will be laggy as hell

Not_That

287 Posts

July 13 2011 13:22 GMT

#20

I want to congratulate the OP for using frames passed between click and action to determine latency. This is the only way that matters. tracert is irrelevant. I hope everyone start using this method to estimate their sc2 latency so that results would actually mean something for discussions.

SecondChance

Australia603 Posts

July 13 2011 13:30 GMT

#21

On July 13 2011 22:22 Not_That wrote:
I want to congratulate the OP for using frames passed between click and action to determine latency. This is the only way that matters. tracert is irrelevant. I hope everyone start using this method to estimate their sc2 latency so that results would actually mean something for discussions.

How is it irrelevant?

Clicking the screen results in the move command being sent as packets of data to the server; whilst the current location of the unit is sent back as it is moving and stopped.

How is any different to a tracert? The tracert simply stamps MS times on each hop the packet takes to get to it's destination.

Also the use of speedtest.net is not really an accurate way of doing it. Who's to say there servers don't experience congestion or latency during your test?

GreEny K

Germany7312 Posts

July 13 2011 13:36 GMT

#22

On July 13 2011 12:33 ehalf wrote:
Interesting. I also live in the US east and my ping to the servers are:
NA:80
EU:150
KR:350

Play on ping with 150 is sometimes noticeable but generally hardly affect the gameplay. But 300ms ping is killing me,

Where in the East? I live in Ohio and I wanted to get an EU account, might be wasting time with 150 ping...

Phaded

Australia579 Posts

July 13 2011 13:39 GMT

#23

Guess I'm not playing on Korea. 360ms cos we go from australia to tokyo to hong kong to korea.

Not_That

287 Posts

July 13 2011 13:47 GMT

#24

On July 13 2011 22:30 SecondChance wrote:

Show nested quote +

It is irrelevant because it is impossible to tracert all the way to the game server. Watching the frames is the only way to know your real latency to the game (which is the only latency that matters).

arbitrageur

Australia1202 Posts

July 13 2011 14:05 GMT

#25

On July 13 2011 22:39 Phaded wrote:
Guess I'm not playing on Korea. 360ms cos we go from australia to tokyo to hong kong to korea.

There have been some posters clamiing they get 200ms to KOR from AUS. Perfectly fine. This is solely due to ISP, I'm guessing. FKN Optus 400ms!!!!

Manaldski

229 Posts

July 13 2011 14:30 GMT

#26

If you are on the US West Coast it should be playable as long as the ping to the korean server is lower then 200 . From Central Europe you can probably get as low as 215 ping and if you tunnel it you can probably lower the ingame delay to about 300ms.

arbitrageur

Australia1202 Posts

July 13 2011 16:15 GMT

#27

On July 13 2011 23:30 Manaldski wrote:
If you are on the US West Coast it should be playable as long as the ping to the korean server is lower then 200 . From Central Europe you can probably get as low as 215 ping and if you tunnel it you can probably lower the ingame delay to about 300ms.

What does tunneling it mean?

bwally

United States670 Posts

July 13 2011 18:26 GMT

#28

On July 13 2011 22:30 SecondChance wrote:

Show nested quote +

Hmmm no, you issue a command on your end and it will only start executing the command once it receives some type of OK. If you watch the video you can obviously see I tell a zergling to move and it won't start moving until it receives the OK back from the server. In this type of RTS engine to save bandwidth only User Input is sent to the server, no unit locations. See this to learn more.

Not_That got it right on "real latency" part but I've explained this like 2 times already. What do I care what the ICMP ping is? What if the game engine adds 100ms+ of latency per command. Since no one knows what the game actually does on the network side the ping only gives you a really rough idea of the latency that is inherent in TCP/UDP, network routing, etc and doesn't account for the game engine's functions. Now if you can reduce ICMP ping to LAN levels and I'll assume like 1ms for now then you can figure out what the engine latency is and you cannot change this so the number is somewhat irrelevant. But over the internet your icmp ping will have a "multiplier" effect on your actual game latency. And if you solely look at the ping to determine whether the latency is acceptable to play you won't have an accurate figure. Because in reality it's much worse with added latency by the game engine.

Speedtest is fine for getting a general ping to the region within a few 100 miles or so. Congestion probably didn't have any effect on my tests, since those "pings" are quite good for me. And if there is congestion the majority of the time then you can't avoid it anyway. But since there is a large difference between Ping and actual game latency it could mean 3 things. Game engine adds latency (which we can guess based on how online synchronous RTS game engines usually work) or battle.net servers are really crap and can't handle the load so their way more congested more than the network around it where I pinged or within 1 hr after doing the ping test the battle.net servers network got congested.

ehalf

408 Posts

July 15 2011 04:28 GMT

#29

On July 13 2011 22:36 GreEny K wrote:

Show nested quote +

Where in the East? I live in Ohio and I wanted to get an EU account, might be wasting time with 150 ping...

I'm in Pittsburgh. I think the most important factor here factor here is which ISP do you use.. I'm a comcast customer and my ping to the EU and KR sucks

Normal

Please or register to reply.

Analysis of NA, EU, KR, Local latency from NA

Completed

Ongoing

Upcoming