Melee Delay Testing

**Byrthnoth** · 2012-12-05 18:58

Delay testing:
Control: 23:58:47 - 9:10:07 : 9470 swings - 183+183 JAs, 0 WSs :: 33080 seconds
With WSs: 9:47:09 - 18:58:29 : 9357 swings - 183+183 JAs, 183 WSs :: 33080 seconds

200 delay weapon, 0% Haste in gear, Single-wielded without any sources of DA/TA. I spaced apart the abilities/weapon skills by 4 seconds to make sure I got the full JA delay from them and used the WS last.

Control Script:

Code:

alias rep No Foot Rise;wait 4;Reverse Flourish;wait 177.5;
alias ten rep;rep;rep;rep;rep;
alias hour ten;ten;ten;ten;ten;ten;
alias fourhour hour;hour;hour;hour;
alias twelvehour fourhour;fourhour;fourhour;

WS Script:

Code:

alias rep No Foot Rise;wait 4;Reverse Flourish;wait 4;energydrain;wait 10;input /echo --tp ?<tp>;wait 163.5;
alias ten rep;rep;rep;rep;rep;
alias hour ten;ten;ten;ten;ten;ten;
alias fourhour hour;hour;hour;hour;
alias twelvehour fourhour;fourhour;fourhour;

Algebra:

Spoiler: show

WS Delay*183 = 22600 Weapon Delay (+/- 200 possible measurement error)
WS Delay = 123.5 Weapon Delay (+/-1.093 possible measurement error)

For now WS Delay with Energy Drain seems to be about 123.5 Weapon Delay. This would work out to approximately 2 seconds.

Reconfirmation of JA delay using control:
200 delay weapon, 9470 swings, 366 JAs used, 33080 seconds
If 58.85 delay = 1 second, we get that each JA used cost us almost 2.5 seconds of delay.
If we calculate it the other way and assume 1 JA = 2 seconds, we get 58.44 delay per second.

Edit: problems mostly solved. Control coming.
</tp>

**Motenten** · 2012-12-06 01:53

Bleh.

Ok, looks like you need a control for the control. IE: another run of the same length with no JAs used at all.

If we assume that JAs have the same delay as weaponskills, and given that 87 swings were removed for 183 weaponskills, 366 JAs should have removed 174 swings.

That gives a possible baseline rate of 9183 swings in 33080 seconds, which is 3.6023 seconds per swing, and 55.52 delay per second.

Of course a curious symmetry arises there: the value is one half the supposed ~110 delay per second that ranged weapons get.

I tried digging out my old zanshin parses to see if I could get a bead on overall delay, but unfortunately don't have anything that can give useful info.

Calculating some limits based on the control parse:

Absolute minimum delay per second: 54.47
This assumes that JAs added no delay whatsoever.

Absolute maximum delay per second: 56.34
This assumes JAs add 3 seconds' worth of delay (significantly higher than any test measurement)

If JAs have a 1.5 second delay, delay per second is 55.39

Doing my own small test --

Lvl 50 Drg/blm, no gear except a Ceremonial Dagger (delay 183)
49 Str vs Ul'hpemde for 0 damage

1234 swings (not including the first one)
12:08:14-1:10:28 = 1:02:14 == 3,734 seconds
3.025931928687 seconds per swing
60.48 delay per second

Restarted parse after turning off uTorrent, in case of network lag.

20:07 = 1207 seconds
399 swings
3.025062656642 s/swing
60.49 delay/second

So my test parses yield yet another value for delay per second, significantly different from yours, as well as different from the 'accepted' 58.85, and even higher than the round 60/second value. On the other hand, it's fairly consistent with itself so far.

I'm wondering if it's getting influenced by CPU clock frequency. I seem to remember that overclocking caused issues for people because of the way the game tied into the system clock. Since my system has a Core i7, it automatically adjusts its maximum frequency based on processor load. Right now it's fluctuating between 60% and 90% of max with a fair bit of regularity (ie: about 15 low/high switches in the last ~minute).

I have no idea how that would impact my swing rate, nor do I have any idea how I could compensate for it during testing if it is.

And of course there's still the uncertainty about if there's a difference between the live server and the test server.

**Motenten** · 2012-12-06 03:20

Longer stretch without torrent running:

6,865 seconds
2,265 swings
3.030905077263 s/swing
60.37800436999 delay/second

So 60.38 delay per second

**Byrthnoth** · 2012-12-06 05:40

Wouldn't it be funny (also known in some lands as "annoying") if a base stat affected delay?

I logged in and confirmed that none of the gear I was wearing had a hidden effect using Reraise. For the record, it was:
Trainee Knife/_/_/_
_/_/Snow Pearl/Snow Pearl
_/AF3+2/Icesoul Ring/Icesoul Ring
_/_/_/_

Also, I had some problems with splitting the parse in Kparser so my endpoints for each sample were single attack rounds from the logs via logger. Doing the control now.

Spoiler: show

Edit: Nevermind. I just realized that the script I wrote to parse the log from logger's output didn't include misses. Now it does and I've adjusted the above post to reflect that.

**Byrthnoth** · 2012-12-06 11:00

Same setup as above, I had two copies of FFXI running (and kparser) running for all tests:
17575 seconds = 5151 hits

3.412 seconds per hit
58.617 delay per second

If I use this to calculate based off my Control above, I get 2.1 seconds per JA almost exactly.

If I use this to calculate weapon delay based off my sample above, I get 2.1 seconds per JA (2.1068...., but variance should be higher).

I don't know what this wonkiness converting between seconds and delay for different delay weapons is, but I do think that JA and WS delay are the same and are 2.1 seconds (~123 delay, which is meaningless to me).

Edit:
If I apply those numbers to here, where I was dual wielding 200 delay knives with 30% Dual Wield:
http://www.ffxiah.com/forum/topic/27...ge/82/#2086373

It appears it's approximately accurate. I backcalculate to 396 base delay assuming no truncation, which is perfectly reasonable with truncation. Haste Samba also works out to 101/1024 Haste again in both of the lower two cases.

**Motenten** · 2012-12-06 11:56

Edit: Nevermind. I just realized that the script I wrote to parse the log from logger's output didn't include misses. Now it does and I've adjusted the above post to reflect that.

Ah, that makes a fair bit of difference. Much better.

Does the swing count include both the first and last swings? If so, the first needs to be dropped for the purposes of calculating total intervals.

Possible rounds: 9469 or 9470
Total time sampled: between 33079 and 33081 seconds

From the first data set:
Swings @200 delay * 9470, JA * 120.0 delay, =>> 58.581-58.585 delay per second
Swings @200 delay * 9469, JA * 120.0 delay, =>> 58.575-58.579 delay per second

Swings @200 delay * 9470, JA * 122.4 delay, =>> 58.608-58.611 delay per second
Swings @200 delay * 9469, JA * 122.4 delay, =>> 58.602-58.605 delay per second

Swings @200 delay * 9470, JA * 123.5 delay, =>> 58.620-58.623 delay per second
Swings @200 delay * 9469, JA * 123.5 delay, =>> 58.614-58.617 delay per second

Swings @200 delay * 9470, JA * 124.6 delay, =>> 58.632-58.635 delay per second
Swings @200 delay * 9469, JA * 124.6 delay, =>> 58.626-58.629 delay per second

Overall delay per second shows a range between 58.575 and 58.635

Going to the control 0 data (no JAs):

5150-5151 swings
17574 - 17576 seconds

delay per second:

5150 @ 17576: 58.603
5150 @ 17574: 58.609
5151 @ 17576: 58.614
5151 @ 17574: 58.621

Since it's within the range given by control set 1, and very close to the control set 0 values, I'd probably pick a flat 58.6 for simplicity, assuming that the reported 5151 swings is 5150 delay intervals.

Extrapolating forward again, that puts the JA delay of control set 1 at between 2.081 and 2.086 seconds, or between 121.9466 and 122.2585 delay.

Data set 2: 122.458 - 122.671 delay per JA when considered in total.

Can you please verify the numbers used for swing counts?

**Byrthnoth** · 2012-12-06 12:02

It does include both the first and last swing. Otherwise the numbers are right. I am running another test with a Ram Staff right now (366 delay) to see if it gives me a different result.

366 delay:
1154+2225-1 = 3378 swings
20832 seconds (5:47:12 Total Fight Length)
= 59.35 delay/second

Slightly lower.
I'd be unsurprised to learn that there's some kind of rounding or tier'd-ness going on when it comes to calculating real attack speed.

**Kirschy** · 2012-12-06 18:57

Originally Posted by Byrthnoth

I'd be unsurprised to learn that there's some kind of rounding or tier'd-ness going on when it comes to calculating real attack speed.

This is always the conclusion I came to when running slow tests a while back. I never was able to hit the accuracy you guys were with the tests you're running, but it seemed to heavily favor some kind of rounding.

Interesting stuffs though~

**Byrthnoth** · 2012-12-07 07:03

Yeah... I'm thinking that these variables need to be eliminated before we can really say it's tiered:
1) Specific computer used. - Seems unlikely to me because SE would have to be borderline retarded to let attacks be initiated client side.
2) Level dependence. - Motenten did his test at level 50. I'm doing mine at 90. We know SE minorly adjusts WS Alpha over the levels, so perhaps they adjust this conversion rate too.
3) Inconsistencies in implementation - It's possible that different weapons have different delays even if the delay in the description is the same, much the way that 5% Haste can take different values (50/1024 or 51/1024).

If it is tiered and doesn't depend on the above things, we could proceed using D1 weapons with different delays to tile the environment. If that's not good enough, we can dual wield weapons (first verifying how Dual Wield works by combining two weapons and a quantity of dual wield so that theoretical round delay = a known weapon delay and calculating delay/second) and get more data points that way.

This is not an insignificant difference, really. Motenten's Ceremonial Dagger attacks 3% faster than you'd expect when comparing it to Trainee Knife. 3% DPS? How many millions of gil is that worth?

So far I see these options:

Code:

                                                                                                                                                                                                                                                                                                                                                                                                                                                         <table border="0" cellpadding="0" cellspacing="0" width="364"><colgroup><col style="mso-width-source:userset;mso-width-alt:4571;width:94pt" width="125"> </colgroup><colgroup><col style="mso-width-source:userset;mso-width-alt:1536;width:32pt" width="42"> </colgroup><colgroup><col style="mso-width-source:userset;mso-width-alt:2523;width:52pt" width="69"> </colgroup><colgroup><col style="width:48pt" span="2" width="64"> </colgroup><tbody><tr style="height:15.0pt" height="20"><td class="xl65" style="height:15.0pt;width:94pt" height="20" width="125">Name</td><td class="xl65" style="width:32pt" width="42">Delay</td><td class="xl65" style="width:52pt" width="69">Delay/Sec</td><td class="xl65" style="width:48pt" width="64">Hits</td><td class="xl65" style="width:48pt" width="64">Seconds</td></tr><tr style="height:15.0pt" height="20"><td class="xl65" style="height:15.0pt" height="20">Ceremonial Dagger</td><td align="right">183</td><td class="xl66" align="right">60.35</td><td align="right">2265</td><td align="right">6865</td></tr><tr style="height:15.0pt" height="20"><td class="xl65" style="height:15.0pt" height="20">Trainee Burin</td><td align="right">190</td><td class="xl66">        59.50
</td><td>       2182
</td><td>       6965
</td></tr><tr style="height:15.0pt" height="20"><td style="height:15.0pt" height="20">Edgeless Knife</td><td align="right">195</td><td class="xl66">
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td class="xl65" style="height:15.0pt" height="20">Trainee Needle/Knife</td><td align="right">200</td><td class="xl66" align="right">58.61</td><td align="right">5151</td><td align="right">17575</td></tr><tr style="height:15.0pt" height="20"><td class="xl65" style="height:15.0pt" height="20">Lady Bell</td><td align="right">216</td><td class="xl66" align="right">59.43</td><td align="right">6970</td><td align="right">25327</td></tr><tr style="height:15.0pt" height="20"><td style="height:15.0pt" height="20">Ibushi Shinai</td><td align="right">228</td><td class="xl66">
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td style="height:15.0pt" height="20">Chipped Scimitar</td><td align="right">236</td><td class="xl66">
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td style="height:15.0pt" height="20">Curtana</td><td align="right">236</td><td class="xl66">
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td class="xl65" style="height:15.0pt" height="20">Trainee Sword</td><td align="right">240</td><td class="xl66">
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td style="height:15.0pt" height="20">Lotus Katana</td><td align="right">280</td><td class="xl66">
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td class="xl65" style="height:15.0pt" height="20">Shell Scepter</td><td align="right">288</td><td class="xl66">
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td style="height:15.0pt" height="20">Club Hammer</td><td align="right">300</td><td class="xl66">
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td class="xl65" style="height:15.0pt" height="20">Trainee Hammer</td><td align="right">324</td><td class="xl66">
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td style="height:15.0pt" height="20">Treat Staff</td><td align="right">356</td><td class="xl66">
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td class="xl65" style="height:15.0pt" height="20">Signet Staff</td><td align="right">366</td><td class="xl66" align="right">59.35</td><td align="right">3379</td><td align="right">20832</td></tr><tr style="height:15.0pt" height="20"><td style="height:15.0pt" height="20">Pitchfork</td><td align="right">396</td><td>
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td style="height:15.0pt" height="20">Caver's Shovel</td><td align="right">402</td><td>
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td style="height:15.0pt" height="20">Dull Claymore</td><td align="right">444</td><td>
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td style="height:15.0pt" height="20">Wooden Katana</td><td align="right">450</td><td>
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td style="height:15.0pt" height="20">Blunt Lance</td><td align="right">492</td><td>
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td class="xl65" style="height:15.0pt" height="20">Trainee Axe</td><td align="right">504</td><td>
</td><td>
</td><td>
</td></tr><tr style="height:15.0pt" height="20"><td style="height:15.0pt" height="20">Relic Weapon</td><td align="right">999</td><td>
</td><td>
</td><td>
</td></tr> </tbody></table>

I do not have access to the Chipped/Blunt weapons. Right now I'm doing Trainee Axe.

Edit: Also, if I'm not mistaken then we have kind of an unreasonable degree of precision in our tests (1 second / N hits = measurement error?). Filling out the table shouldn't really take that long, should it?

Edit2: So far I'm noticing ~58.6, ~59.4, and ~60.4 as the values. However, the precision on the values clustered around 59.4 should exclude each other unless we have another source of variability.

**Prothescar** · 2012-12-07 07:52

Is it possible that weapon delay is similar to, say, haste, in that it's based off of a fraction and thus, while a dagger may say it has the same delay as another one, it may have a slightly higher/lower delay than its counterpart due to rounding? Ceremonial and Trainee are pretty far apart implementation wise, could be they used different values when making them. Maybe you've already accounted for this and I'm dumb for suggesting it, idk.

**Byrthnoth** · 2012-12-07 07:55

That's also possible. I'll add it as a third option to be eliminated. The most obvious candidates for testing it are Signet Staff vs. Ram Staff. I used Ram Staff to test 366, but Signet Staff was implemented years before.

**Motenten** · 2012-12-07 13:58

Doing a comparison run to see if the difference is related to the computer used.

Weapon: Signet Staff (366 delay)
Job: Drg/blm, lvl 50

Note: I found the error with my original data. I did the hit count by counting the number of lines pulled from the raw data, since I was using that to get the timestamps. However I forgot that it included "scores a critical hit!" lines, so I counted more lines than there were actual hits. Calculations based on that higher number yielded a delay per second of ~60.5.

Did a verification against Offense page this time, compared with the Raw Data with the critical hit lines taken out. Everything matched up. New calculations:

[12:29:06 PM] 15,bc,78,80707070,000015e4,0000199a,0021,00,01,02, 00,Motenten misses the Ul'hpemde.1
...
[1:37:36 PM] 15,57,77,80707070,0000187e,00001c47,0021,00,01,02, 00,Motenten misses the Ul'hpemde.1

666 attack intervals (667 hits if you include the first one)
68:30 = 4,110 seconds
6.171171171171 seconds per hit
59.30802919708 delay per second

Uncertainty range:
59.29 - 59.32 delay per second

Pretty close to your 59.35 average on Signet (or Ram?) Staff data.

Went back to correct the data from the Ceremonial Dagger parses:

[1:18:29 AM] 15,39,8a,80707070,000005ab,000005e9,0021,00,01,02, 00,Motenten misses the Ul'hpemde.1
...
[3:12:54 AM] 15,97,88,80707070,00000e5c,00000ec2,0021,00,01,02, 00,Motenten misses the Ul'hpemde.1

2,223 attack intervals (2,224 hits, verified)
1:54:25 = 6,865 seconds
3.088169140801 seconds per hit
59.25841223598 delay per second

Uncertainty range:
59.25 - 59.27 delay per second

That puts everything but the Trainee Needle between 59.25 and 59.50, with 3 of the 5 samples being between 59.25 and 59.35.

**Byrthnoth** · 2012-12-07 14:34

The Trainee Axe came in in that range as well (1713 hits in 14516 seconds -> 59.44 delay per second).

I went back to check the Trainee Knife parse and made sure the numbers were correct, and they were. I guess I'll leave it on tonight. It's possible that the Hpmede walked away from me and then wandered back or something, which would have given me lower hits and lower delay/second than expected.

**Byrthnoth** · 2012-12-08 09:28

I ended up doing Trainee Sword instead (240 delay) and it came up at 58.9 delay/second (5563 hits, 22660 seconds). I'm not seeing any strong trend so far:

Today I think I will take a break from this and work on /ra crit rates, per this post:
http://ffxi.allakhazam.com/forum.htm...186510825877#1

I have myself on 90 DNC with Mamoolbane /ra-ing the lowest level Ul'hpmede in sea for 0 damage (I have 77 AGI). After I get a few thousand ranged attacks in, I'll take it to the highest level Om'hpmede to determine if ranged crit rate relies on dAnything.

**Kirschy** · 2012-12-10 15:46

Originally Posted by Byrthnoth

I ended up doing Trainee Sword instead (240 delay) and it came up at 58.9 delay/second (5563 hits, 22660 seconds). I'm not seeing any strong trend so far:

Today I think I will take a break from this and work on /ra crit rates, per this post:
http://ffxi.allakhazam.com/forum.htm...186510825877#1

I have myself on 90 DNC with Mamoolbane /ra-ing the lowest level Ul'hpmede in sea for 0 damage (I have 77 AGI). After I get a few thousand ranged attacks in, I'll take it to the highest level Om'hpmede to determine if ranged crit rate relies on dAnything.

Reminds me of an interesting test Nayla ran a while ago. She was doing ranged attacks on the mobs outside Bastok and was coming up with a RA Crit Rate of around 18% even with a ton of AGI+ on, and 5% in crit merits. I don't have the time right now, but I'll see if I can find her data later. It seemed like there was a unique formula to ranged crits.

Thread: Melee Delay Testing

Thread Tools

Melee Delay Testing

Melee Delay Testing

Latest Threads

Up & Coming Threads

Hottest Threads