• Binkenstein

It looks like it varies between 1.8 and 1.9, so having a variance of roughly 5.5% is probably not going to be a good thing when 3.0 could be between 2.835 & 3.165

 

Rouncer

please elaborate.

 

Nite_Moogle

Dump all talents. Use a 1.5 speed weapon. Cast windfury on it. Go kick the shit out of a target dummy with auto attack for an hour with only auto-attack. If the cooldown is exactly 3 seconds, you should always see two swings between "adjacent" procs. That is, you should see:

white hit (0)
WF Hits (indeterminate)
white hit (1.5)
white hit (3.0)
white hit (4.5)
WF hits (indeterminate)

If you see the following listing happen with exactly 1.5 speed weapons, the cooldown is slightly less than 3 seconds and tiny amounts of haste should be added until the above happens.

white hit (0)
WF Hits (indeterminate)
white hit (1.51)
white hit (3.02)
WF hits (indeterminate)

The weapon speed at which you see this happen is the cooldown of windfury. You will probably need to use a 1.6 speed dagger with increasing amounts of haste to prove the exact cooldown, or a more significant range in which it occurs.

This works because auto-attack is a toggle state, and you only send start and end to the server. Once you are auto-attacking the server knows when your next auto-attack will land and does not wait for the client to tell it when the next auto-attack happens.

 

Xoya

Split Greathammer and The Hand of Nerub are both 1.6 speed weapons that most shamans have probably picked up already. I'll work on this a bit after raid tonight if it hasn't been done prior to that.

 

Tramana

Using [Necrolord's Sacrificial Dagger], naked otherwise, no talents. windfury on the blade. 3.0 seconds between WF procs. 27ms latency. On the PTR.

Switch to [Jambiya]. 2.8 seconds between WF procs.
I think we can conclude that it's not a 3.0 second cooldown anymore (if ever?). I'll continue testing.

edit:
Switch to [Poniard]. After 53 wf procs (106 total wf hits) I found none that were 2.6 seconds apart.

Switch back to [Jambiya]. add 111 haste rating, 3.39% haste. Tooltip says 1.35 second swing. WF proc 2.7 seconds apart.
Using 186 haste (5.67%). Tooltip says 1.33 second swing. After 50 wf procs (100 total wf hits) I found none that were 1.66 seconds apart. I suspect 1.7 is the number,

Using 126 haste with a 1.4 second dagger that should put me just under 2.7, I still got 2 wf 2.7 seconds apart.
So the cooldown is somewhere between 2.66 and 2.7 seconds.

I ran with 129 haste, and after 42 windfury I got two within 3 swings. I ran with 132 haste, and got no close windfury procs over 52 wf. I ran one more time with 129 haste and didn't get any close windfury procs. Maybe the one close one I got was a server fluke?

Tested with 0 haste again and after 50 wf, no close winfury procs. I'm guessing at this point that either I made a mistake earlier, or the PTR isn't reliable for this testing.

 

Nite_Moogle

Even though the attacks that triggered WF were less than 3 seconds apart, the WF procs themselves were at least 3 seconds apart (at least according to the log). Very interesting.

 

Rouncer

Still for our purposes that means that the "windfury chasm" isn't going to occur with weapon's hasted down to 1.5 seconds, it's going to occur when you get enough haste to bring the swing time under 1.35 seconds.

Tramana, thank you for all that work. That's simply amazing, I really wasn't expecting anything near these results so quickly.

I'm going to try and confirm your findings on Friday. When they are confirmed we can start working on making a chart of the exact amounts of haste required at each weapon speed to hit that chasm for the TTT thread. Once confirmed we can also give the info to Tukez so that when he updates the sim for the new Windfury glyph he can also update the sim for the correct windfury cooldown as well.

 

Tramana

That's an interesting observation. Perhaps the proc check is lagged a bit, and explains the irregular results I was getting. As it got later in the evening < 3.0 second procs went away.

It would be interesting to see if this can be reproduced on live servers.

 

◊ Wraithlin

Because clearly everything happens server side in exact the right time, which is why lag never affects your... oh wait. As moogle pointed out, the WF procs are still stamped as being 3seconds apart even if the auto-swings are stamped as being <3.0s apart.

Possible explanations are:
1) For some inexplicable reason WF is coded to hold onto the proc for up to 0.5seconds
2) The time stamps aren't reliable indicators of when a WF proc was generated.

Some of those WF attacks are stamped at 1.1s after the autoattack which generated them, that is a pretty long delay in processing. The issue is we don't know how the server resolves WF procs in such a way that they can appear over a second after they occur, so we cannot interpret the timestamps meaningfully.

None of which precludes the CD being 2.7s!
And this is the problem, with unreliable time-stamps it is pretty hard to prove anything.

You have to do more than prove that 2.7s swing timers can proc WF, you have to show that, within statistical noise, the number of WF procs as a percentage of auto-attack swings does not decrease when moving from 1.5 to 1.35s swings. The existence of a apparently sub 3.0s procs is inconclusive, because we don't know how WF is resolved nor how time-stamps are generated.

[e]
Another possibility is that haste lowers the WF CD in the same way it lowers the GCD.

[ee]
Basically, I think we need more data before we can assert anything. The variation induced by lag is too large, just looking at the data provided here the resolution of WF is stamped at anywhere from 0.5 to 1.1 seconds after the generating auto-attack.

Further, to simplify things significantly, we should be testing this with only 1 weapon equipped (2h or a single 1h). The interaction of two swing-timers (and possibly two WF imbues?) is only going to make this even harder to determine.

 

Rouncer

I don't think haste is playing a role since for most of the tests shown here haste wasn't involved. Also even with an incredible amount of haste (WF totem + Flurry + 10% passive haste + 2xmongoose) the cooldown appears to be right around 3 seconds (+/- 0.3s).

The other thing is whether it is important or not that the windfury attacks are being delayed at some point. The issue is that when you pass a certain haste point you lower your weapon speed to a point where it requires an additional swing before it is eligible to proc a windfury so even if the procs are delayed if the swings are still eligible then it shouldn't really matter.

 

◊ Wraithlin

Well, it matters but in a way that is not as obvious as "when you drop below this number you see a disjoint drop in WF damage".

Assume that there is some delay between auto attacks and the generation WF attacks, and that delay is normally distributed. Then the time between the generation of WF procs is also normally distributed even with a static swing timer. So what you will observe as your weapon swings drop below the hypothetical CD is a gradual reduction in WF procs because some tail of your WF distribution will still exist above the CD.

Ok so I added a graphic.
Imagine the top is the distribution of "swing timers" as recieved by that part of the WF code which checks for wether WF is on CD. The lower graphic is what you get with faster weapons, some of the differences will still creep up to being large enough that WF can proc, there is no "sharp edge". This roughly agrees with what we have seen empirically (i.e your DPS won't dive if you take a little bit too much haste).

AFAIK, the only place we have ever "observed" sharp drops in DPS (which is what should be seen if you have just enough haste to cross a break-point) are in the output from simulations which assume instant resolution of WF with auto-attack.

Attached Thumbnails

 

 

MatsT

That might not even be enough. If the reason we can procc wf from autoattacks closer than 3 seconds or even get wf proccs closer than 3 seconds is latency and lag on the client and server, there is not really any way to measure the 'wf cooldown'. In fact, you could argue that there is no static cooldown of it. In the end it means that the black spots for haste is more or less removed, but it's very hard to simulate properly.

 

Exewut

Can we assume WF behaves in the same way when it's on a 2h than when it's imbued on only one hand while dual wielding? If we use that assumption then we should be able to test this rather easily. If we use a 3,0 speed weapon without any haste then the # of wf procs should be 40% of the total swings. Now just move to a 2,9 weapon and see if this behaviour continues. If the CD is really 3s then the expected number of WF swings should be around 2*(,8*,2+,2*,0) = 32% (it's been a while, but I believe this is the correct formula, either way the % of wf swings would be noticeable lower at the point where your weapon speed crosses the cd time). Just repeat the same process with base haste pushing a weapon that would normally be slower than the cd under the cd to see if haste affects the CD.
There's no need at all to rely on timestamps from the combat log this way.

edit: there is actually a 3,0 speed onehanded weapon, so you don't even need to use the assumption that 2h works in the same way as 1h imbues. [Kovork's Rattle]

 

Rouncer

The thing is that we don't have just a little bit of haste right now, we have a ton of haste so suddenly we are seeing hitting at that point consistently not just when the shaman is lusted or has a haste proc active.

What I am concerned about is that all the previous testing with haste and interference from the WF cooldown were made with the assumption that the WF cooldown was 3 seconds and also that the shaman was using WF/WF. We are using WF/FT now so the interference is that much more profound and if it does turn out that the cooldown is intend less then 3 seconds the previous testing that showed that the haste chasm did not exist in actual fact was flawed and does not apply towards our current situation.

Personally, I just want them to dump the notion of a cooldown and instead change WF to a PPM mechanic with an inability to occur on 2 melee hits in a row but since we play the game as it is not how it could be that is a moot point.

So the other thing that needs to be tested is to determine if there really is a sharp drop in overall dps at specific weapon speeds. Tramana would you be willing to test that? Use the 1.5 speed weapon and track number of procs over a decent interval and then repeat with the 1.4 and then the 1.3 and see if the "cooldown" does cause that sharp decline that Wrathlin brings up.

 

◊ Wraithlin

If there is any distribution of delay between auto-attack and windfury generation which is not very close to a delta-function (i.e. almost no "spread") then there is no such thing as a "black spot". And this agrees with what is seen empirically, there is no haste "chasm".

Basically, if the swing timers were perfect, and WF were perfectly correlated to them, then there should be some attack speed at which adding 1 point of haste causes a lot of "clipping" which is observed as a large drop in damage. The gain from 1 point in haste is negligible so such a "step" should be pretty obvious. However if WF procs occur with some reasonable spread, then no such point would ever occur, because a shift in the mean of the distribution would cause a continuous, rather than disjoint, drop in WF procs.

This is something we can test, we simply need to know the number of WF procs which occur back-to-back with various weapon speeds around the 3.0 mark. Pick up a 3.1 2-hander and plot the %age of WF attacks followed by another WF attack against haste rating. If there is no lag and a hard CD then we should see a sharp change around the CD speed from 0% to 20% of WF attacks being followed by a second. If we see a sigmoid curve of reasonable width in the haste domain then we have a lag-moderated distribution. In either case when the plot intersects 0.1, you have the middle of the distribution and you can determine the CD. We could even determine the approximate variation of the distribution.


I'd be happy to help collect some of this data, but my shaman is level 70 and I am unlikely to change that any time soon.

A reasonable sample would be 200WF procs (So about an hour) at each swing speed, from 3.1 down to 2.6, in relatively small step sizes (approximately 0.02s in the swing time domain would work). That should not be too difficult with a reasonable range of weapons, just slow!

[e]
We would need to search the log for events which appear as: WF- Swing-WF.

 

Tramana

Can't do much more testing until friday night. Given that I couldn't get "swing, wf, swing, swing, wf" later in the night with a 1.4 speed and no haste, there's likely a drop off. I like the bell curve illustration. I suspect that the bell curve around the cooldown grows or shrinks depending on server load.

Another thing I found while testing is that the wf proc appears to be around 20% with the windfury weapon glyph on the PTR. I tested with a 3.4 speed two hander, for about 400 swings, and got around 82 procs (164 hits). If this isn't a reported bug it'd be nice if someone could confirm and report it to blizzard.

 

◊ Wraithlin

I suspect this is true, however, we an add together data with different variances (widths) as long as they have the same mean; and this is what we are assuming when we talk about a fixed WF CD. We are only interested in finding the mean for the moment.

I am happy to help with the log/data analysis if people can provide combat logs, I am just a little challenged on the "level 80 shaman" front!

 

Mengus

Regarding WF proc delays and such, the behavior is very evident in pvp, where I've died, but had my WF proc go off and kill someone post death...

My combat log has looked like the following on multiple occasions:

Stormstrike/random white hit
You have died.
Your Windfury Attack hits so and so for X
Your Windfury Attack hits so and so for Y
So and so has died.


So whether this is purely a latency issue or client/server interaction, it's definitely a factor that makes accurate assessment of the 3s cooldown very difficult at best.

 

Lujaar

Actually there is a blue post that uses the words "3 second cooldown." Blue Post: Tseric vs Quadfury

Of course, CMs have been wrong before and Tseric was infamous for making nonsensical statements on the class boards. ("Resto shamans have less mana than other classes because when they go oom they can melee the boss.")

 

Paenor

As I was one of the people quoted from talking about FT/FT earlier, I'll weight in on your results here.

First off, i'll agree with you that Recount is doing something screwy. If you look at the proc count that recount gives, against reading it out of the combat log you will find they normally aren't the same.

Over the past 2 weeks i've been doing further testing, and have seen data which matches with both hypothesis. When testing on target dummies with rank10/rank10, I quite often am losing FT procs. However, I have had a few tests where I don't lose them (Just did one now with 175 melee, and 173 FT with 1.8/1.4 speed weapons for example).
Additionally, I've seen bosses where I've had no lost procs ((example), and some where I have gained over 100 extra (example).

Now for a little background on those results. My server is one that suffers hugely from the lag that you might have seen Blizzard mention patching on the PTR. The Maexxna fight was done on a Tuesday, when the server wants to grind to a halt and your abilities have a 8 second lag to them. Which tends to bring me to the conclusion that much of the issue comes from some sort of server processing issue. This is echoed by the fact that you tend to get FT procs in bursts (or after you stop attacking), rather than after each attack. However, the fact that using rank 10/9 seems to fix some of these issues, makes the issue a little more confusing.

Back on topic:

I just threw WF/WF onto my regular gear and went at a target dummy for 5 minutes. 144 haste rating, 1.8/1.4 mongoose/mongoose. I had around 170 ping. (Normally I have ~300 during prime time on the dummies. I can get down to like 60 but only if it's 4am and i'm out in the middle Desolace or somewhere equally in the middle of no where. I blame the server.)

1/15 10:24:23.812 SPELL_DAMAGE,"Windfury Attack",
1/15 10:24:23.812 SPELL_DAMAGE,"Windfury Attack",
1/15 10:24:26.421 SPELL_DAMAGE,"Windfury Attack",
1/15 10:24:26.421 SPELL_DAMAGE,"Windfury Attack",

2.609


1/15 10:27:52.125 SPELL_DAMAGE,"Windfury Attack",
1/15 10:27:52.125 SPELL_DAMAGE,"Windfury Attack",
1/15 10:27:54.656 SPELL_DAMAGE,"Windfury Attack",
1/15 10:27:54.656 SPELL_DAMAGE,"Windfury Attack",

2.531

The first took place during a regular rotation, and the second during a heroism timed with my whetstone and dueling totem procced so I was attacking at ~0.7/0.5. I'd definitely say that there isn't a three second cooldown.

 

Squidfury

Temporary spikes can really throw results off. If the packet signalling a WF proc were delayed by 700ms on the first proc, and the server sent the second proc exactly 3 seconds later, but it only took 200 ms to reach you, there would only be a 2.5 second difference in the combat log. That seems pretty feasible( or even likely), especially on a laggy server.

Thats why time stamps have no authority. Latency fluctuates wildly over small periods of time. Ping the server a few times to take a snapshot rather than an average to see what I mean.

The weapon speed test is the only viable test. It doesnt necessarily prove that the CD is less than 3 seconds either. Who knows how it interacts with thier netcode? However, a weapon speed test does show us that practical purposes there is no dead zone at 1.5 seconds. At the very least there are some that happen sooner that 3.0, and very unlikely that any happen before 1.35. As previously stated.

BTW, it would only take a latency spike of 1000+ to cause weapons this fast to show a WF proc caused by one weapon swing, occur after the following weapon swing on your combat log. This would invalidate even the weapon swing test. You would have to insure that you never recieve spikes of latency up to weapon speed X100.

edit: I cant ping a server from work, but I pinged yahoo 10 times and got results up to 200 ms apart. I would imagine the blizzard servers tend to have less stable latency.

 

SentinelBorg

Some time ago I wrote a small program to parse the combat log and search for wf procs closer then 3 secs. It did find only 2 cases in logs of a months raiding.

 

• Caladiera

Hey Rounced,

What length of combatlog would be most beneficial? I have some free time and good latency (~30ms) so I can batter the test dummies for a while.

weapons from 1.3 -> 2.5?

 

• Binkenstein

If you're going to spend some time on it, ideally I'd go for as large a data set as possible. 10,000 hits per weapon would be ideal, but 1000+ should be sufficient.

 

Bellante

An input that might seem a bit off-topic when you read the first line, but bear with me.

I seem to recall someone wanting to do the achievement in storm peaks to hit 10 mines without landing, and he succeeded by, oddly enough, opening 20 youtube videos on a fairly slow connection so his ping went haywire, and somehow the interaction with the server seemed to get his character able to just run through 10 mines (with 1 fps) without getting blown off the ground. Others had problems reproducing his result, either because it didn't work, or because their connection was able to handle the simultaneous download, who knows.

Related to this issue, would a similar approach, buffing your latency tremendously, be a viable method for testing how much latency affects the server-client issue? Perhaps be able to actually produce a combat log that shows windfury procs on white hits in a row with a speed of, say, less than 2.5 or so? Anyway, just a thought for inspiration.