seku

(excuse me for my English)

If you were to prove the cause is lag. Could it be possible, (with a proper stochastic model of server lag) to improve your dps?
Does someone have some data about WoW servers' lag modelisation?

Does someone have some data on tanked mobs (no parry)?
Is it possible to get some lag data with these?

 

• Disquette

Hey all, after my first tests, when Slant pointed out that parry was affecting the results, I've specifically left in *all* parry entries, and typically don't use them in the example data sets I post because they can change things around.

And yes Vulajin, my other attachments in page two of this thread also show similarly weird behavior.

Now, on to the most recent tests (dual 1.4 speed weapons). I'm not sure if lag is involved or not.

Oh, and before the screenshots, I must say I'm always impressed when someone writes "excuse me for my English" and then puts together a better constructed post than most American high school kids would :-)

So your question, seku... Yes, I think by combing lag with a high hit rate, you could actually cheat the system, but WoW wouldn't be very enjoyable with the amount it would take to achieve a meaningful dps boost. It's also interesting to me that people using fast weapons probably don't realize that one of the reasons they're not as far behind in dps tests is because of lag >.<

Ok, so the pics and analysis...

Pic 1 is a very standard eventlog readout - one crit and 4 attacks within the flurry boundaries.

Occasionally, however, when lag is bad, you get results like screenshot 2, where you have a single crit and *six* attacks are within the flurry boundaries.

Pic 3 is an example of 7 flurried attacks happening after a crit/flurry string.

I *think* this data is meaningful, but it's hard to say since the combat log isn't useful for showing weapons as flurried or not - lag makes it too difficult to say when a 1.4 speed weapon is flurried based on combat log parsing. Also, I think a legitimate case could be made that having same-speed weapons influences the outcome. I don't know that for sure, however, and would have to do yet another test ( /cry).

So, if weapons half the speed of the former weapons test were being used, and we didn't get 2x the number of flurried attacks (or even close to that amount), then lag isn't (the only) answer. I don't know what to do now.

(btw, at first I thought this was lag, but i went back over the data from previous posts, and realized i was just seeing what I *expected* to see - Trying to be objective, I'm now thinking that this looks a lot like the other tests).

Attached Thumbnails

   

 

 

Xerophyte

If the lag is only flurry fading later than it should and not cumulative with earlier charges then the effect wouldn't be so extreme as 2x the number of Flurried hits, you'd only see an increased frequency of 3+ streaks, with more streaks of 5-6 hits within a non-chain flurry than with slow weapons. However, your fraps-sync test would appear to mean we can discard that explanation. Which sorta puts me at a loss ... it's annoying when your pet theory gets shot down.

Is there any more information to be gleaned from the client than what the combat log says? Do any events fire when Flurry charges change? My wow lua chops are fairly weak, but if I understand correctly you can call UnitBuff() and get the current stack number of your Flurry. If stack changes fire the UNIT_AURA or PLAYER_AURAS_CHANGED events it should be possible to catch them to track when Flurry charges do what they do... I think. Any one with good knowledge on the innards of the game feel like chiming in on viability here?

 

The Grog

It is possible that these results are due to fluctuating lag?

For example, if the client receives 'Flurry Charge Consumed' messages from the server abnormally close together due to shifting lag, that it might discard one or more of them as echos/resend/general screwups?

That could explain the larger odd flurries with fast weapons, as it would be easier for charge consumed messages to get lagged on top of each other.

 

• Disquette

The Grog. Here's something I've *never seen* in my testing so far - 2 flurry charges being used (as shown on screen) with no hits in between. Now, that could be luck, or it could be more meaningful. The more tests I do, fast weapons, slow weapons, whatever, without seeing that, makes me pretty certain that what you're saying is about right - When the client says "process a reduction in flurry charges", it won't do more than one at a time, and it will wait till the server has said "ok, I hear you, you've reduced it by one" before it does another.

 

seku

May you emulate some 1000+ ms lag while holding some fast weapons+haste to find out if this behaviour is systematic? (and exploitable :P)
I'm sorry I can't do much, I'm currently enduring exams.

 

Amorpheus

What do you reckon are good weapon speeds for flurry? Or does it not matter that much? I'm planning to get the season 2 arena weapons but the offhands are all 1.5 speed... I'd like something faster than "main hand speed" but 1.5 looks like it'll waste more charges on offhand than necessary. Fist weapons are 2.6/1.8 though...

Let's assume the 2.6 MH isn't hasted but the 1.5 OH is - the offhand will hit twice between mainhand hits and that just seems terrible to me. With a 1.8 OH in the worst case they would still alternate. Am I worrying about nothing here or will this impact flurry significantly?

"You are better than I am," Inigo admitted.
"So it seems. But if that is true, then why are you smiling?"
"Because," Inigo answered, "I know something you don't know."
"And what is that?" asked the man in black.
"I'm not left-handed."

 

D4vE

Just a suggestion:

Grab a tank to help you testing and perform all attacks from behind to avoid the influence of parry on your data. At least it would rule out the "a parry hasted attack does not consume a flurry charge" theory, if you get more than 4 hasted attacks after a crit.

 

Pater

For the benefit of the warriors in this thread, I'm cross-posting a link to a small parsing mod I wrote to test Flurry charge consumption.

http://wowinterface.com/downloads/in...FlurryLog.html

It's a workaround that uses the say channel and chat (not combat) logging to output you hits, misses, and flurry charges. Not suitable for use near anyone else because it spams the say channel, but perfect for Blasted Lands servants.

 

• Disquette

Yeah, it'd be nice. my main problem is lack of people willing to test. In order to set up a 5 person test, I had to literally pay 100 gold per person for a 30 min blasted lands session. I do have a pally who said he'd help in the future though, so that could work.

 

Wandre

Wow you guys jump really far really fast. I can assume a great testing group would have a few rogues and a warrior/pally tanking with rogues using slice and dice from behind so no parries got in the way.

If you'd like to test without worrying about mobs but instead using player tests why not just go into a PVP area and get a mage to cast slow or a warrior to use thunderclap. I think you could at least get some good data on weather or not a swing can be slowed in mid swing.

Grab a nice 3.60 speed 2H wep and have a war thunderclap you. You should be able to time things easier with the slow 3.6 and you can do it for hours without worrying about mob death or respawns.

 

Maratai

Just use a Bear Tank. No parry.

 

Freyalis

Hi Guys,

I'm just trying to iron out some facts about flurry both for my own purposes and to add to my collection of Enhancement theorycraft files I'm building rapidly on my PC.

We Know that:

* Flurry reduces the attack speed of your weapons by a factor of 1.3
* Attacks that have been launched Prior to a Flurry proc ie. Mid swing are not hasted.
* Any auto attack that is launched while flurry is active has its speed hasted.
* Attacks that have been launched with the flurry haste are not de-hasted if flurry wears off while the attacks are in mid swing.

* A Crit with any white or Yellow melee attack will proc / refresh flurry.
* Yellow melee attacks do not consume flurry charges.
* Regular melee hits, glancing blows, and blocked hits all consume a charge.
Q. Does a hit that is fully blocked (ie 0 dmg) still consume a charge?
* Missed hits do not consume charges.
Q. Does a parried hit consume a charge? I would think not as is doesn't land

Known flurry bugs:

* When weapon speed is syncronised in a lag free environment, usually created by pre-engaging auto-attack before entering melee range whilst using 2 equal speed weapons, 4 flurried attacks are generated with the 3 charges. (This is fairly obvious to see given the rules for when attacks are hasted as written above)

Unknown flurry bugs:

* Flurry strings of irregular lengths have been tracked in combat logs, anywhere from 2 - 7 flurried swings using 2 different speed weapons, and 2-4 swings using only 1 weapon.
* Lag is effecting the consumption of flurry charges, particularly when 2 attacks land very close together.


In regards to the questions above i would love to just dive into game and test them myself right now but my game card ran out last week, and due to my wedding coming up in 3 weeks time i wont be playing for a little while =P.

 

• Disquette

The only new information I have at the moment is really making me cringe. I tested 6 different weapon speed configurations for 12minutes+ each. These were:

2.6/2.6, 2.6/2.0, 2.6/1.4, 2.0/2.0, 2.0/1.4, and 1.4/1.4

The only combination which showed an excessive number of extra attacks, given my crit rate, compared to the number predicted by using the flurry closed-form formula was the 1.4/1.4

That is, all other were within 1.2% of the predicted value using 1-(1-CritRate)^3 as the chance for an attack to be flurried.

I'm not sure how you came up with the known fact of:
Maybe to explain my confusion with this fact, I can post some combat logs of attacks showing that even 2.6/2.6 weapons get way out of sync in combat logs. I just don't buy the equal-speed bonus yet. When I can post from my other computer which has this weekend's testing data, you'll see what I mean.

 

dukes

Are you sure that the reason they get out of sync is because of flurry? Looking towards the middle of this thread, there were some comments made about parry speeding up attack speed, which I assume only affects MH swings, and that could be throwing them out of wack if you were solo attacking one of the BL servants.

 

• Disquette

pure combatlog parsing with no parries also shows it. I'm on my other computer now working with the lag model. I think you're going to be surprised with the data when I tablulate and post it.

It's comparing a simulation with no lag jitter against a simulation with lots of lag jitter versus the closed-form theory.

 

Ailee

I've noticed that with auto shot, the haste effect does not apply until the following auto shot. I'm guessing it would work the same or similar to auto attacks. Food for thought.

Edit: It does also seem to apply the opposite way; If an attack starts before the haste is removed, it'll carry through until that one is finished.

 

• Disquette

Yes, that part I think we can all agree on - with the exception of parrying an incoming attack, there are very few things that haste one of your attacks mid swing. The rogue that gave the example of SnD earlier gave one such instance of confirmation of this.

 

• Disquette

I've attached the data I was talking about.

Sim850. Lag is worked into the model as:
Next swing = Lag/2 - Random()*Lag. (So it averages out at regular hit rate)
Flurry = grace period if swing happens within Lag range.
Lag = 850 milliseconds.

Please correct this part if you think I've done this wrong.
Theory. For this row, theory says number of hits =
Total non-flurry-talent attacks + (30% * FlurryUptime% * NonFlurryTalentAttacks)
FlurryUptime % = 1 - (1-Crit)^3

Example with 20% crit, 2600 seconds, and dual wielding 2.6 speed weapons.

nonFlurryTalentAttacks = (2600/2.6) * 2 = 1000 * 2 = 2000
FlurryUptime = 1 - .8^3 = .488
so TheoryFlurryAttacks = 2000 + (30% * 48.8% * 2000) = 2000 + 293 = 2293.

Also of note - I'm guessing that the test data for the last column, the 2.0/2.0 is probably a bit flukey considering how off the % extra attacks is compared to all the others. If it were at the same extra attack rate as the others (about 20%), it would also match the Sim850 results within 0.2% or so. Of course, that needs to be tested, and me just jawboning about it doesn't do much for actual data needs :-)

Attached Thumbnails

 

 

Pater

I take it this was without the WF Weapon imbue?

 

• Disquette

Correct - no enchants, no haste weapons. Plain jane grey/white weapons and flurry only.

 

Freyalis

Very Interesting,

I wont be giving up on the BS + 2.6 OH combo just yet then.


However I am still curious about whether parries, dodges, partial and full blocks consume flurry charges.

My gut instinct says no for parry, dodge and yes for partial block, which is functionally identical to a crushing blow. As for full Blocks, I have no idea but its not really important as you should never see this happen by the level you can spec into flurry.

I only ask as the somewhat dated entry for flurry in wowwiki mentions that parry/dodge/block all consume charges and that was the only information I could find on it.

 

Rezarel

A minor correction: what you call FlurryUptime is the chance that a single swing will be flurried. It is not the fraction of time you spend in the flurried state.

Some math to show the point, similar to your sample calculation (edited to just one 2.6 speed weapon to simplify things):

Flurried attacks = 48.8% * 1146 = 559
Nonflurried attacks = 51.2% * 1146 = 587

Time spent flurried = 559 * 2.6 / 1.3 = 1118
Time spent nonflurried = 587 * 2.6 = 1526

Total time = 1118 + 1526 = 2644 != 2600 as originally assumed.


To correct for this:
FC = Flurry Chance -- what you were calling FlurryUptime
WS = Weapon Speed

Flurry Uptime Fraction = (Time spent flurried) / (total time)
Flurry Uptime Fraction = (FC * WS / 1.3) / ((1 - FC) * WS + FC * WS / 1.3)
Flurry Uptime Fraction = (FC / 1.3) / ((1 - FC) + FC / 1.3)

Flurry Uptime Fraction = 1 / (1.3 / FC - 0.3)
Flurry Uptime Fraction = FC / (1.3 - 0.3 * FC)


So, for Flurry Chance = 48.8%, Flurry Uptime Fraction = 42.3%
TheoryFlurryAttacks should be 2000 + (30% * 42.3% * 2000) = 2000 + 254 = 2254.

 

• Disquette

I have to run now to go to 4th festivities, but wanted to respond saying thank you, and that I figured something was wrong, but not sure what. I appreciate the post, and will review it so that I understand it more fully tonight.

Well my friend is taking a bit longer to get ready, so I had more time. I agree with your analysis and explanation - thank you! In reality it only makes things look better for the simulation instead of the closed-form, simply because the closed form already under-represents # of attacks currently.

I do hope that Tornhoof sees this set of posts so that he can let us know how he deals with this in his closed form method of flurry calculation.

Thanks again.

 

galzohar

Regarding speeds, remember that while a fast offhand will consume more flurry charges, it will also proc more flurries. So overall I doubt there's a real difference between the flurry uptime for main hand when you change the speed of your offhand and vice versa. Just like 2X1.5s weapons would have the same flurry uptime as 2X3s weapons, I bet the same could be figured out for weapons of different speed - intuitively flurry uptime for each one weapon should be independant of the speeds of the weapons. Of course there may be some wierd mechanics I'm missing here so it's worth looking into, but it definitely doesn't let you assume that "fast OH consumes charges => fast OH is bad".