[Enh] Windfury Weapon Cooldown - Page 3

• Caladiera · 01/16/09, 9:28 PM

Attached is a combatlog of me + 1.3 dagger on a lvl60 combat dummy.

Talents: 0/0/0
items: [Plans: Deadly Bronze Poniard] + tabard :p
Location: Thunder Bluff Expert's Training Dummy (lvl60)
Skill: 400/400
latency: 21-25ms
time: 55min (until poniard hit durability of 0)

I'll start on the rest when the server calms down again :/

Rouncer · 01/16/09, 11:09 PM

WWS of that 1.3 dagger.

Wow Web Stats

Looking at that combat log is it possible that the windfury is being applied by the server before the melee swing that has been determined to be proccing it? Because there are numerous timestamps of windfuries 0.1seconds before the next melee swing.

Malan · 01/16/09, 11:34 PM

Sounds like we need to figure out what's being determined server and client side first since that seems to be the discrepancy.

Binkenstein · 01/17/09, 12:29 AM

May be useful to get a script to count the number of normal hits between WF hit pairs.

Broshious · 01/17/09, 3:38 AM

Was just doing some testing myself with a 2.9 two-hander and I noticed a back-to-back proc which shouldn't be possible with a 3 second CD.
Further 2.9 data: (I know it's not enough but it's what I've got for now =P)

853 attacks
324 WF attacks = 162 WF procs

If the CD is 3 seconds then after each WF proc there is one attack that cannot be a WF proc leaving 691 attacks able to proc WF which would give 691*.2 = 138 WF procs.

If the CD is 2.9 seconds or less then all 853 attacks are eligible to proc WF giving 170WF procs.
Forgot to add that my latency is ~90ms so perhaps that can account for the discrepancy in numbers.

Taking the 1.3 second WWS data:

2338 attacks
664 WF attacks = 332 WF procs

If the CD is between 2.6 and 3.9 seconds then there are 2*332=664 attacks that cannot be a WF proc leaving 1674 attacks able to proc WF which would give 1674*.2 = 334 WF procs which seems to be the case.

The 1.4 data:

1066 attacks
333(?) WF attacks (assuming one WF attack got missed somehow =/) = 167 WF procs

If the CD is between 2.8 and 4.2 seconds then there are 334 attacks that cannot be a WF proc leaving 732 attacks able to proc WF which would give 146 WF procs.

If the CD is less than 2.8 then there are 167 attacks that cannot be a WF proc leaving 899 attacks able to proc WF which would give 179 WF procs.

Did a quick little bit of 2.8 testing:

1327 attacks
440 WF attacks = 220 WF procs

If CD 2.8 or more then 220 attacks can't proc WF so 1107 can. 1107*.2 = 221 estimated procs
If CD less than 2.8 = 265WF procs

• Caladiera · 01/17/09, 3:52 AM

Attached is a combatlog of me + 1.4 fist on a lvl60 combat dummy.

Talents: 0/0/0
items: [Right-Handed Brass Knuckles] + tabard :p
Location: Thunder Bluff Expert's Training Dummy (lvl60)
Skill: 400/400
latency: 25ms
time: 27min (was asked to heal a successful Undying run woot)

This one barely gets over the 1000 hit threshold.

Wow Web Stats

Atren · 01/17/09, 5:08 AM

This came up once long ago as well and was dealt with as some disparancy with how it is processed. You can have WF procs in same second even depending on something. I can't remember what it was, but if the megathread still around somewhere you should be able to find the exact reasoning it was given then.

Ok, not sure if it is same type of question:

http://elitistjerks.com/651809-post7356.html

Broshious · 01/17/09, 5:43 AM

Originally Posted by Atren

This came up once long ago as well and was dealt with as some disparancy with how it is processed. You can have WF procs in same second even depending on something. I can't remember what it was, but if the megathread still around somewhere you should be able to find the exact reasoning it was given then.

Ok, not sure if it is same type of question:

http://elitistjerks.com/651809-post7356.html

He says he had procs 2.74 seconds apart explained by his 26ms latency, but 26ms is .026 seconds not .26 seconds as he suggests.

Rouncer · 01/17/09, 9:34 AM

WWS parse of the 1.4 fist data

Wow Web Stats

So to put everything together what we really should be looking for is a drop off in the (number of windfurys)/(number of normal melee hits) at a specific weapon speed with a large enough sample size to give the results statistical viability.

On the PTR I'm currently testing with a 2.9 mace hasted down to 2.67 at the moment. After an hour (1000 hits) I'm going to remove enough haste to take it to just over 2.7 speed and repeat and then again at 2.8 and then finally again naked with it at 2.9 speed. If the cooldown is 3.0 then there should be no variation between the ratios at any of those speeds. Anyone see anything wrong with that logic?

(e) damn, talk about statistical variability, just had 20 hits with a 2-hander between windfury procs and that's with the new windfury glyph. 33 between just now, maybe our mechanics are still a little too RNG?

Atren · 01/17/09, 10:58 AM

RNG creates variations. You can have it even with same speed weapons so there is no reason to assume there is none with different speeds. Rounced, are you also removing hit rating during that test? Since white attacks proc WF any reduction can influence it further.

Concerning the link I posted in my earlier post, the picture is the main focus of it. It shows 0.72 seconds between windfury procs unless I misunderstood.

Rouncer · 01/17/09, 12:07 PM

Originally Posted by Atren

RNG creates variations. You can have it even with same speed weapons so there is no reason to assume there is none with different speeds. Rounced, are you also removing hit rating during that test? Since white attacks proc WF any reduction can influence it further.

Concerning the link I posted in my earlier post, the picture is the main focus of it. It shows 0.72 seconds between windfury procs unless I misunderstood.

Hit Rating plays no role since I'm using the level 60 dummies so there are no misses and no dodges or parries.

That picture, if I am understanding it correctly, is showing that the WF procs are seemingly determined server side and that through lagging out his client he was able to get 2 wf procs to essentially happen back to back because his client received both from the server basically at the same moment due it being lagged out from his manipulations.

Basically there should be a speed where the cooldown no longer plays a role in proc rates. So the proc rate should theoretically be 25% on the PTR with the new WF glyph when you reach that point.

I think I will run a baseline set with a 3.8 speed weapon, after this series concludes, and see if the proc rate does indeed match the null hypothesis (being that without interference from the cooldown the proc rate should be 25% with the new glyph in place).

Atren · 01/17/09, 12:59 PM

Ok, sorry for this post as i had DW penalty in my mind, but this test does not require DW

• Caladiera · 01/17/09, 1:12 PM

1.5 speed fist

Talents: 0/19/52 - resto (no melee haste related talents)
items: [Right-Handed Claw]
Location: Thunder Bluff Expert's Training Dummy (lvl60)
Skill: 400/400
latency: 22ms
time: 55min

Around 43min in my toon lost contact with the test dummy for 2 swings (hungry animals :/). It is past the 1000 swing threshold so the data can be truncated at that point.

Wow Web Stats

Wraithlin · 01/17/09, 3:29 PM

Were you using the WF glyph?
Anyway, here is my parser which shall be exposed in all gory detail so any mistakes are spotted:

    Do
        Line Input #1, Line
        Marker = Mid$(Line, 20, 12)
        If Marker = "SPELL_DAMAGE" Then
            WFProcs = WFProcs + 1
            If SwingCount = 2 Then
                TrailingProcs = TrailingProcs + 1
            End If
            SwingCount = 0
        ElseIf Marker = "SWING_DAMAGE" Then
            SwingCount = SwingCount + 1
        End If
    Loop Until EOF(1)
    
    Print #100, ""
    Print #100, "WF Procs      :", (WFProcs / 2)
    Print #100, "Trailing procs:", TrailingProcs
    Print #100, "Follow-on %age:", (TrailingProcs * (200) / WFProcs)

Results:

File 1:       15fist.txt
WF Procs      :              376 
Trailing procs:              31 
Follow-on %age:              8.24468085106383

File 1:       14fist.txt
WF Procs      :              180 
Trailing procs:              6 
Follow-on %age:              3.33333333333333

File 1:       13DaggerWFLog.txt
WF Procs      :              370.5 
Trailing procs:              14 
Follow-on %age:              3.77867746288799

The data sets are small, but the "lag bell-curve" hypothesis is looking pretty reasonable so far. If anything the suggestion is that the cut-off on WF may, in fact, even be slightly higher than 3.0, but we really need a lot more data.

• Caladiera · 01/17/09, 5:29 PM

Originally Posted by Wraithlin

Were you using the WF glyph?

All my tests were done on live, so yes I do, but its not the increase chance to proc one on ptr

[e] Anyone have an idea how to get around durability loss? I could run longer data pulls, but even epic weapons will fall apart after about an hour and a half.

Rouncer · 01/17/09, 5:36 PM

I've been watching myself hit a dummy all morning with various slow weapons and here are the results including a little additional testing checking on the new WF glyph and I have a theory on how this is all working.

I think the cooldown really is 3 seconds but the discrepancies comes about because of the client/server interactions. I think the determination of whether it is a WF proc occurs client side and then gets sent to the server where it is determined if the WF cooldown was active or not and then the server sends it back to the client informing it whether the WF actually occurred or not.

So for an autoattack - the server keeps the time on the auto-attacks based on positioning and what not and when the data hits the client it determines if it was a windfury proc or not which then gets relayed back to the server to determine if it was on cooldown and then the proc info goes back to the client where it is reported. For an instant attack whether it is a proc or not seems to be sent along with the info about the attack right to the server so there is less lag associated with it.

These were all on the PTR with roughly a 100-125ms latency and with the new Windfury Glyph in place.

2.67 speed - 17.2%
Wow Web Stats

2.71 speed - 17.8%
Wow Web Stats

2.9 speed - 19.1%
Wow Web Stats

3.0 speed - 20.79%
Wow Web Stats

3.08 speed - 19.5%
Wow Web Stats

3.3 speed - 24.96%
Wow Web Stats

So basically if you are using a weapon speed less then a division of 3.0 + latency (possibly even 2xlatency) you are occasionally running into the cooldown and losing some potential procs but on the other side if you are more then a division of 3.0 - latency (possibly 2xlatency) you are sometimes gaining procs which gives that bell curve appearance to the proc rates that Wrathlin was talking about.

Here are two short (only 300 hits each) tests with a 3.8 speed which seem to show that the new glyph is working as intended.

With glyph - 25.67%
Wow Web Stats

Without glyph - 19%
Wow Web Stats

Finally here is a quick parse of just some stormstrikes without any autoattacks
Wow Web Stats

Notice that the delay between the stormstrike hit and the windfury is roughly 0.8 seconds while for autoattacks it is more like 1.2 seconds. It requires more testing but it seems to me to show that for the instants whether it is a windfury proc or not is determined client side and relayed to the server while for autoattacks they are relayed to the client and then determined and then relayed back to determine if it actually occurred or was blocked by the cooldown and then finally was reported to the client/combatlog.

Wraithlin · 01/17/09, 5:48 PM

I would be surprised if WF procs are declared clientside, but the WF check my be called from the client after checking if the WF buff is active. Something like:

Server > Client Autoattack result
Client > Server WF buff is active, check if it proced
Server > Client result of WF check.

The interesting part of this result is quite subtle though. If the CD is 3 seconds and there is some "spread" on this due to lag, when you are at the 3.0 CD you actually lose about half your back-to-back procs. The "sweet spot" for not losing procs would be around 3.2seconds, which is what rounced's data appears to show

[e]
Rounced, any chance you can send/post those logs?
Generating a second parse to pull out only your actions should only take a few minutes.

Rouncer · 01/17/09, 5:58 PM

Originally Posted by Wraithlin

[e]
Rounced, any chance you can send/post those logs?
Generating a second parse to pull out only your actions should only take a few minutes.

Here ya go.

Atren · 01/17/09, 6:51 PM

Wraithlin, I really doubt Blizzard would trust client computer with so important task. It is too risky for hackers to take advantage of such design.

Wraithlin · 01/17/09, 6:52 PM

I had to edit the parser for the slower swing speeds, but here are the results

File 1:       rouncer_267.txt
Swings :      1

WF Procs      :              210 
Trailing procs:              0 
Follow-on %age:              0

File 1:       rouncer_271.txt
Swings :      1

WF Procs      :              184 
Trailing procs:              2 
Follow-on %age:              1.08695652173913

File 1:       rouncer_281.txt
Swings :      1

WF Procs      :              183 
Trailing procs:              1 
Follow-on %age:              0.546448087431694

File 1:       rouncer_29.txt
Swings :      1

WF Procs      :              193 
Trailing procs:              14 
Follow-on %age:              7.25388601036269

File 1:       rouncer_30.txt
Swings :      1

WF Procs      :              220 
Trailing procs:              27 
Follow-on %age:              12.2727272727273

File 1:       rouncer_308.txt
Swings :      1

WF Procs      :              195 
Trailing procs:              30 
Follow-on %age:              15.3846153846154

File 1:       rouncer_33.txt
Swings :      1

WF Procs      :              194 
Trailing procs:              62 
Follow-on %age:              31.9587628865979

File 1:       rouncer_38.txt
Swings :      1

WF Procs      :              77 
Trailing procs:              18 
Follow-on %age:              23.3766233766234

File 1:       rouncer_38_NG.txt
Swings :      1

WF Procs      :              57 
Trailing procs:              13 
Follow-on %age:              22.8070175438597

Wraithlin · 01/17/09, 6:55 PM

Originally Posted by Atren

Wraithlin, I really doubt Blizzard would trust client computer with so important task. It is too risky for hackers to take advantage of such design.

Probably true, but part of the problem is exactly this. We really don't know for sure what is generated where.

[e]
With the PTR glyph, we would expect a follow on %age of 12.5 if our weapon speed was equal to the CD on WF, so a number near to 3.0 still looks most likely. However that does not change the fact that you would want to have your swings landing at around 3.2s to avoid "lag clipping" of WF procs.

The sweet spots would be 3.2, 1.6, 1.1, 0.8.

Protico · 01/17/09, 9:19 PM

It seems to me that figuring out what is calculated server side and client side etc. is the long way around. In the end we want to find out if attack speed will affect windfury DPS, why not just hit the dummy a few thousand times with a 1.4, 1.5 and 1.6 speed weapon with equal DPS and see which gives the highest DPS. If current theories are true (3 second cooldown) 1.4 speed will be markedly less DPS than 1.5, which will be marginally better than 1.6.

Wraithlin · 01/18/09, 5:59 AM

Two reasons.
First you bring in extra variables (misses, crits).
Second you cannot combine data from many people, because of variables such as AP, str, agi, arPen, armor on target ....

Counting WF procs is the best way to remove all the variables except the WF CD and the WF proc chance.

Protico · 01/18/09, 11:12 AM

Originally Posted by Wraithlin

Two reasons.
First you bring in extra variables (misses, crits).
Second you cannot combine data from many people, because of variables such as AP, str, agi, arPen, armor on target ....

Counting WF procs is the best way to remove all the variables except the WF CD and the WF proc chance.

If all that changes is the weapon than over a decent number of examples misses/crits (really only crits because we don't miss the level 60 dummy much) get smoothed out just the same way that RNG gets smoothed out for the counting method. Also I wasn't thinking about combining data, just having one person swing at the dummy a lot.

The reason I suggest this method is that it doesn't seem like counting procs will give a definitive answer until MANY MANY swings down the road.

Wraithlin · 01/18/09, 11:56 AM

If you think that 1.5 should out perform 1.6, you have, in fact, missed the point.

Also this work is only a small part of the big picture; trying to extrapolate the effect of weapon speed on DPS from how it influences your DPS when naked, talentless, and auto-attacking a target dummy is an act of futility. The purpose is not to work out how weapon speed affects DPS, but how weapon speed affects WF proc's so that it WF be better modelled by shaman simulators. Then those simulators can be inspected to compare the DPS of various weapon and armour combinations.