|
Recorded at 60 FPS with a mutalisk. All data is delay between green circle showing up and command card lighting up measured in frames. Each frame is 16.(6) ms apart (60 FPS). Encoder settings: crf=0 which should make it lossless, but I'm not sure if I were getting every keyframe.
8 Turn Rate Low Latency
9, 11, 13, 13, 11, 13, 10, 11, 11, 12,
12, 12, 16, 11, 11, 12, 10, 14, 10, 10
average: 193ms
12 Turn Rate Low Latency
6, 7, 7, 6, 5, 7, 5, 5, 2, 6,
8, 4, 5, 4, 3, 6, 8, 6, 10, 5
average: 96 ms
12 Turn Rate High Latency
9, 7, 10, 8, 10, 9, 10, 5, 7, 10,
9, 4, 7, 10, 10, 9, 9, 10, 7, 9
average: 141 ms
12 Turn Rate Extra High Latency
9, 10, 13, 12, 12, 12, 12, 15, 13, 12,
11, 13, 10, 10, 9, 10, 10, 11, 12, 12
average: 190 ms
Conclusions:
It seems turn rate 8 is similar to turn rate 12 extra high latency. Not sure why, but in my tests turn rate 12 low latency was twice as fast.
Bonus meme:
numbers for iccup:
11, 7, 12, 12, 14, 11, 13, 9, 8, 13,
13, 12, 14, 11, 9, 15, 9, 13, 13, 10
average is 11.45 which is 191ms, roughly equivalent to Turn Rate 8 Low Latency, or Turn Rate 12 Extra High
|
you know I think it would be great if turn rate 12 was set after a ping test before start of game on ladder, don't need turn rate 12 all the time on but having it on when possible would be great
so according to your data if ping test returns less than 100ms set to turn rate 12 :D?
|
Nice, thanks for the data!
|
What is the standard turn rate for the SCR ladder?
|
On August 31 2017 05:12 Gorgonoth wrote:
What is the standard turn rate for the SCR ladder?
AFAIK turn rate 8, but I didn't test it
|
|
|
TLADT24920 Posts
On August 31 2017 05:12 Gorgonoth wrote:
What is the standard turn rate for the SCR ladder?
It's TR8
|
On August 31 2017 05:07 ProMeTheus112 wrote:
you know I think it would be great if turn rate 12 was set after a ping test before start of game on ladder, don't need turn rate 12 all the time on but having it on when possible would be great
so according to your data if ping test returns less than 100ms set to turn rate 12 :D?
that's not what the data says, actually
it implies this is how it works:
if you have TR8, you will get a response between 125 and 250 ms depending on when you clicked
if you clicked right before a sync point, you'll get 125 ms, if you clicked right after, you'll get 250 ms
average is then 187.5 ms which seems to fit the data - but remember, if your ping is MORE than 125 ms, you'll sometimes lag
TR12 would then be between 83.(3) and 166.(6) which averages to 125 ms, but you need less than 83 ms ping not to lag with that person
although for some reason my average is way faster on TR12 than I would have expected
|
tec27 said TR8 is similar to #LL #L2 on 1.16 (albeit a bit slower) and TR12 on low is much faster than #LL (1.16 lan latency).
I wonder if TR9 // TR10 could be made available in the future, lat speeds would be closer to #LL on 1.16 and it might not lag as much as TR12.
|
On August 31 2017 07:33 TT1 wrote:
tec27 said TR8 is similar to #LL #L2 on 1.16 (albeit a bit slower) and TR12 on low is much faster than #LL (1.16 lan latency).
I wonder if TR9 // TR10 could be made available in the future, lat speeds would be closer to #LL on 1.16 and it might not lag as much as TR12.
he measured it wrong, it takes a few frames for a unit to respond, you need to look at the command card, not the unit moving
he's off by like 50 ms; it's completely wrong
|
You should show data for #L2. Everyone else I've seen test it has come out with TR8 being pretty close to #L2.
|
Netherlands4520 Posts
On August 31 2017 05:07 ProMeTheus112 wrote:
you know I think it would be great if turn rate 12 was set after a ping test before start of game on ladder, don't need turn rate 12 all the time on but having it on when possible would be great
so according to your data if ping test returns less than 100ms set to turn rate 12 :D?
The problem with turn rate being decided before each game is that as a player, I predict having a different turn rate each game will just be confusing for how to actually micro.
The clicks you have to make to execute micro tricks are quite different tr8, tr12, tr16, and playing on tr8 the brain automatically gets used to how to micro on tr8, so playing on different turn rates constantly might make consistency in micro really difficult.
But maybe it's possible for you to simply learn 3 different micro methods depending on what turn rate you're playing on...time will tell!
|
On August 31 2017 07:53 Liquid`Ret wrote:Show nested quote +On August 31 2017 05:07 ProMeTheus112 wrote:
you know I think it would be great if turn rate 12 was set after a ping test before start of game on ladder, don't need turn rate 12 all the time on but having it on when possible would be great
so according to your data if ping test returns less than 100ms set to turn rate 12 :D? The problem with turn rate being decided before each game is that as a player, I predict having a different turn rate each game will just be confusing for how to actually micro. The clicks you have to make to execute micro tricks are quite different tr8, tr12, tr16, and playing on tr8 the brain automatically gets used to how to micro on tr8, so playing on different turn rates constantly might make consistency in micro really difficult. But maybe it's possible for you to simply learn 3 different micro methods depending on what turn rate you're playing on...time will tell!
makes sense for muta micro vs t, where if you miss your accuracy the difference is huge, but as P user I think I dont mind some slightly different lats;; tbh 8 feels good enough but better is great like luxury maybe^^
I found the difference annoying in the past between the high lat on public bnet and "lan latency", but between T8 or T12 I think it's closer just faster is better when possible~
in some rare cases, one might still play on actual LAN :O as they do in offline tourneys
|
On August 31 2017 07:33 TT1 wrote:
tec27 said TR8 is similar to #LL #L2 on 1.16 (albeit a bit slower) and TR12 on low is much faster than #LL (1.16 lan latency).
I wonder if TR9 // TR10 could be made available in the future, lat speeds would be closer to #LL on 1.16 and it might not lag as much as TR12.
The data supports them being the same, not "a bit slower". There is not enough granularity in possible network turn sending rates to account for the differences between the various 1.16 lan latency tests and TR8, which leads to a few possible causes:
- Slightly higher input/graphical lag
- Variance (i.e. more data points would equalize them)
Neither indicates a difference between TR8 and 1.16 lan latency, but the former might help account for the "feeling" of higher latency.
On August 31 2017 07:38 iopq wrote:Show nested quote +On August 31 2017 07:33 TT1 wrote:
tec27 said TR8 is similar to #LL #L2 on 1.16 (albeit a bit slower) and TR12 on low is much faster than #LL (1.16 lan latency).
I wonder if TR9 // TR10 could be made available in the future, lat speeds would be closer to #LL on 1.16 and it might not lag as much as TR12. he measured it wrong, it takes a few frames for a unit to respond, you need to look at the command card, not the unit moving he's off by like 50 ms; it's completely wrong
Literally doesn't matter, since I measured it the same way across all turn rates and game versions. Yes, the command card will respond sooner, but at no point was I trying to measure the exact "to-the-millisecond" latency of the command card. We wanted a relative measurement of 1.18+ turn rates to 1.16 lan latency, and my method gives that (as would yours, most likely, if you had also tested on 1.16, but I somewhat mistrust the command card as it has known issues that cause it to desync from game state).
On August 31 2017 07:53 Liquid`Ret wrote:Show nested quote +On August 31 2017 05:07 ProMeTheus112 wrote:
you know I think it would be great if turn rate 12 was set after a ping test before start of game on ladder, don't need turn rate 12 all the time on but having it on when possible would be great
so according to your data if ping test returns less than 100ms set to turn rate 12 :D? The problem with turn rate being decided before each game is that as a player, I predict having a different turn rate each game will just be confusing for how to actually micro. The clicks you have to make to execute micro tricks are quite different tr8, tr12, tr16, and playing on tr8 the brain automatically gets used to how to micro on tr8, so playing on different turn rates constantly might make consistency in micro really difficult. But maybe it's possible for you to simply learn 3 different micro methods depending on what turn rate you're playing on...time will tell!
Agreed on it being difficult to transition between the 3 rates. I generally think it's a bad thing that Blizzard exposed this as an option to users, and I think its an even worse thing that they made LAN enforce TR16 now. Do they just expect pros to get used to that? To only play TR16 on bnet from now on? I *really* don't understand the thinking there.
|
Has there even been a single word on Server instead of p2p? Could this still be a thing?
|
On August 31 2017 09:31 tec27 wrote:
Literally doesn't matter, since I measured it the same way across all turn rates and game versions.
You had like 4 data points and you had high latency being faster than low, which from my data is absolutely false.
|
On August 31 2017 10:44 iopq wrote:Show nested quote +On August 31 2017 09:31 tec27 wrote:
Literally doesn't matter, since I measured it the same way across all turn rates and game versions. You had like 4 data points and you had high latency being faster than low, which from my data is absolutely false.
Only the case of the average, which should be a good indication to you that average is *not* actually a good way to measure these things since the measurements are pretty noisy. The min/max's all line up as you'd expect (and the deltas between those basically match what you have here)
|
Actually, the average is pretty stable. Take the L8 numbers.
The average is 11.6 - but if you take only the first 10 numbers the average is 11.4
for L12 it's 5.75 and 5.6 respectively
for high latency it's 8.45 and 8.5
for extra high latency it's 11.4 and 12
this is why I bothered to sample twenty times so that I get pretty accurate numbers, instead of only sampling 4 times
numbers for iccup:
11, 7, 12, 12, 14, 11, 13, 9, 8, 13,
13, 12, 14, 11, 9, 15, 9, 13, 13, 10
average is 11.45 which is 191ms
this corresponds to 8 Turn Rate low latency or 12 turn rate extra high latency
|
I think with the TR12 ladder latency change I need to bump this because there are a lot of people who don't understand what TR12 actually means.
|
Since it seems like the delays are indeed the same for 1.16 and 1.20 on tr8, is it possible that the input delay increased with the patches so it overall feels like the combined delay is higher than before? Is it possible to measure the input delay for both cases?
|
On September 18 2017 18:15 Cryoc wrote:
Since it seems like the delays are indeed the same for 1.16 and 1.20 on tr8, is it possible that the input delay increased with the patches so it overall feels like the combined delay is higher than before? Is it possible to measure the input delay for both cases?
I tried muta micro in a custom game on iccup and it felt about the same as 1.20 TR8.
For some reason people overhype "lan latency" on iccup. It's not that amazingly low. TR12 and TR16 is actually much much faster.
|
no but there is a difference from iccup latency and chaos launcher ,i think icc changed or delayed it even more.when i first tried fish server it was faster than iccup.and on fish then the low latency hack took place so yeah.. low latency hack speed is the one from single player,is even faster than normal lan.
|
On September 19 2017 00:10 iopq wrote:Show nested quote +On September 18 2017 18:15 Cryoc wrote:
Since it seems like the delays are indeed the same for 1.16 and 1.20 on tr8, is it possible that the input delay increased with the patches so it overall feels like the combined delay is higher than before? Is it possible to measure the input delay for both cases? I tried muta micro in a custom game on iccup and it felt about the same as 1.20 TR8. For some reason people overhype "lan latency" on iccup. It's not that amazingly low. TR12 and TR16 is actually much much faster.
You are using iccup lan to test your data? smh.. use wlauncher
|
On September 18 2017 18:15 Cryoc wrote:
Since it seems like the delays are indeed the same for 1.16 and 1.20 on tr8, is it possible that the input delay increased with the patches so it overall feels like the combined delay is higher than before? Is it possible to measure the input delay for both cases?
It is possible, yes, and it is *somewhat* possible to measure. I have data for this for various 1.16 configurations measured using Typometer, someone could probably re-measure and compare against 1.20. I'll note two things with that, though: first, it's not possible to measure full-screen configurations with that tool, as it can't "see" the game window contents in that case (and full-screen configurations are likely the lowest latency configuration). Second, W-Mode is actually really slow in that regard, and I'd be surprised if Blizzard managed to do worse than it.
|
I have just played against korean and the game was completely fine, smooth and no lag. After the game I got matched with the same guy - game was totally unplayable even on extra high settings. wtf.
|
On September 18 2017 17:33 iopq wrote:
I think with the TR12 ladder latency change I need to bump this because there are a lot of people who don't understand what TR12 actually means.
Nice bumb. Makes me more confident about my suggestion in here.
By the way, did you also check the reaction times of tr8 on high and extra high latency? I'm trying to understand how the net code currently handles pings higher than 250 ms.
|
On September 19 2017 02:36 Netto. wrote:
I have just played against korean and the game was completely fine, smooth and no lag. After the game I got matched with the same guy - game was totally unplayable even on extra high settings. wtf.
I had that yesterday. Except the first game we played laggy as fuck, 2nd game was insanely smooth. Makes no sense to me.
|
On September 19 2017 00:16 [sc1f]eonzerg wrote:
no but there is a difference from iccup latency and chaos launcher ,i think icc changed or delayed it even more.when i first tried fish server it was faster than iccup.and on fish then the low latency hack took place so yeah.. low latency hack speed is the one from single player,is even faster than normal lan.
AFAIK they use exactly the same add-on under the hood
|
the koreans on bnet forum still writing "why 12 plz make 16" are so anoying they not even try to understand the problem
|
so is there any update coming? unplayable ladder for me so far, im getting to many games vs koreans... (to laggy even on extra)
|
|
|
|
|
|
EPT
