Elitist Jerks Shadowpriest FAQ for 4.0

11/08/10, 10:29 AM   #136
draxius
Glass Joe

Human Priest

Stormrage
You're right, I was a little messy ^^
I did some tests and I wanted to predict the future ones in order to make them quicker.

The board contains both predicted numbers and tested numbers.
The tested numbers are Bold and/or Red/Green.

Here's my new results for S0/1 D1/3 :
Three results were 1 rating point lower and the last one is 1 rating point higher than expected :

 Conditions DP1 SWP1 VT1 DP2 SWP2 VT2 DP3 SWP3 DP4 VT3 DP5 SWP4 DP6 VT4 S0/1 D0/1 205 273 328 615 820 984 1025 1367 1435 1640 1845 1913 2255 2296 S0/1 D1/3 171 238 292 576 779 942 983 1321 1388 1591 1794 1862 2200 2240 S0/1 D2/3 137 204 258 539 740 900 941 1276 1342 1543 1744 1811 2146 2186 S0/1 D3/3 104 170 223 502 701 860 900 1231 1298 1497 1696 1762 2093 2133 S1/1 D1/3 7 71 122 393 586 741 780 1102 1166 1359 1552 1617 1939 1978 S1/1 D2/3 -26 38 89 357 548 701 740 1059 1122 1314 1505 1569 1888 1926 S1/1 D3/3 -57 6 56 322 511 663 701 1017 1080 1269 1459 1522 1838 1876

I removed the S1/1 D1/3 line because it's impossible to test ^^
A couple of suggested modifications to your formula - let me know if these help:
1) I use a required rating of 32.785 Haste Rating per % of Haste in most of my calculations. Tends to work for me.
2) Rounding errors bit me for a while. Try doing it this way:
Haste from Rating (HfR) = 1 + Haste Rating/3278.5
Darkness Haste (DH) = 1 + (Ranks of Darkness/100)
True Haste (H) = HfR * SH * DH
DoT tick Interval (I) = ROUND(3/H; 3) - Round to the 3rd decimal place
DoT tick Count (C) = ROUND(DoT Duration/I)

My predicted Haste Ratings at 80 for additional Ticks (below 1900 haste)
VT: 56, 663, 1268, 1876
SW:P: 6, 511, 1017, 1522
DP: -57, 322, 701, 1081, 1457, 1838

It should be noted that the bolded numbers above have been tested and confirmed as the cut-offs for new ticks by various folks.

I'd also note that the adjusted formula I use predicts the deviations for the S0/1, D1/3 tests that you found.

In the end, if appears that what WoW does internally is:
Calculate all values of haste without roundinging (or with insignificant rounding)
Determines the tick interval of the DoT by rounding to the 3rd decimal
Calculates DoT duration based on the rounded tick interval

I hope this helps folks for future calculations.

11/08/10, 5:49 PM   #137
Bristine
Glass Joe

Human Priest

Drenden
 Originally Posted by Naphomci From my understanding, if DoTs auto-updated in the manner you are saying, it would potentially be a dps loss. You would have cases where you could lose a major benefit to a DoT (i.e. you re-apply DP at the end of heroism/bloodlust and so have 24 seconds of the speedy DP). I think in the end it would probably bring down the dps of shadow priest who maximize effects like that. As for the Dark Archangel: The idea would be to consume right after applying VT/DP so that those DoTs keep the Evangelism buff, and you have enough time to rebuild the Evangelism buff for reapplying VT/DP
That's wrong. Heroism is 40 seconds, and since we don't control when it's popped, we lose up to ~24 seconds of speedy DoT ticks when it goes up after we've applied it. We lose DPS re-applying as soon as it goes up instead of the alternative: letting them tick at hasted speeds while DPSing as normal. Trinkets, too must be macro'd into VT. DP can't be cast on mobs that are disease immune and so that's out, SWP is only cast once per enemy, so that's out. If VT isn't recast often enough in a fight then we might lose buffed uptime. (This is an extreme for very long fights, but it might matter.) We cannot use "chance on hit" trinkets to full effect because they might pop just after we've applied our DoTs and only apply to the following MF's, SWD and a MB.

It is an enormous hinderance to Shadow Priests to not have auto-updating DoT's, we are limited in trinket selection and use and our synergy with group members is hampered as well since we don't effectively gain their buffs as well as a direct damage nuking class would.

11/08/10, 10:30 PM   #138
moowalk
Don Flamenco

Troll Priest

Khaz'goroth
 Originally Posted by Bristine It is an enormous hinderance to Shadow Priests to not have auto-updating DoT's
Having DoTs which don't auto-update is a large increase to shadowpriest dps. It's certainly not worse, since randomly timed bloodlusts (and other effects) will mean an equal amount of lost ticks at the start of the bloodlust as those gained after bloodlust finishes.

Most raids will pop Bloodlust in a predictable manner for exactly the sorts of reasons being discussed. It enables people to time activated dps increases efficiently.

Yes, it might be annoying tracking trinket procs and haste buffs, but if done effectively, it is a dps increase.

Personally, Bloodlust happens in our raids about 9-10 seconds into every fight. I've found that holding off on berserk and then refreshing dots with every single haste buff results in higher initial dps. Doing this, it's possible to refresh VT 3 times and DP twice during bloodlust. This gives 45 seconds of bloodlusted VT and 48 seconds of bloodlusted DP despite bloodlust only lasting 40 seconds. The same reasoning allows 24 seconds of berserked, gloved and potted DP and 15 seconds of berserked, gloved and potted VT, despite berserk, gloves and potion only lasting 10, 12 and 15 seconds respectively.

 I'm pretty confident that it [SW:P] would gain Bloodlust haste as soon as it's refreshed while under the haste buff, and it'll probably drop the buff shortly after BL ends.
Read the thread. If that's too tedious, search for my earlier posts.

SW:P doesn't update haste buffs when refreshed via mindflay. This is probably a bug and may be fixed, but for now, there is nothing more amazing than rolling a SW:P throughout a fight with 5 different haste buffs.

11/09/10, 8:50 AM   #139
Elimbras
Don Flamenco

Dwarf Priest

Eitrigg (EU)
 Originally Posted by Bristine That's wrong. Heroism is 40 seconds, and since we don't control when it's popped, we lose up to ~24 seconds of speedy DoT ticks when it goes up after we've applied it. We lose DPS re-applying as soon as it goes up instead of the alternative: letting them tick at hasted speeds while DPSing as normal.
I'm sorry if this seems rude, but your statement is plainly wrong from a mathematical point of view.
Assume that BL is popped at a random time. Assume also that you refresh DP exactly every 24s. If dots are auto-updated on buffs, then you gain 40s of BL on your DP. Now, let's see what gain you have if dots are updated on cast. Let X denote the time left on your current DP when BL is cast, and let t=0 be (by convention, up to a shift) the time where BL is cast. At time X, you cast a new DP, hasted with BL. The next DP cast is at X+24. If X was less than 16, we have that X+24 <= 40, and this second DP is also hasted by BL. Otherwise, we have X+24 > 40, and the second DP is not hasted.

Now, BL is cast at a random time, independently of DP cast, and DP is cast exactly every 24s. This means that the remaining time X on the current DP is a uniform random variable on ]0, 24]. So, we have 2/3 chance of having X less than 16, and 1/3 of having X greater than 16. The mean gain of BL on DP is hence 1/3 * 24 + 2/3 * 48 = 40s of hasted DP. That's identical to the case of auto-update. In other words, when there is no auto-update, the potential gain after the end of the buff is exactly the potential loss before the first recast of the dot.

There is no magic value of durations here. This will be always valid, as long as the buff is casted at a random time, independently of the dot casting time. Now, if there is no independence between both, then you could potentially be stucked in a bad synchronization. But with an "equivalent"[*] probability, you could be stucked in a good synchronization. And in fact, you have control of it, and you can ensure in most cases that the synchronization is better that the mean random case.
[*]: I don't mean here the probability of being in a good or bad synchronization are equal. I mean that the expected gain, ie. the expectation of the probability of a synchronization times the gain (or loss) of this synchronization, is fundamentally null.

11/10/10, 2:10 PM   #140
Derrall
Glass Joe

Troll Priest

Tichondrius
 Originally Posted by Elimbras I'm sorry if this seems rude, but your statement is plainly wrong from a mathematical point of view. Assume that BL is popped at a random time. Assume also that you refresh DP exactly every 24s. If dots are auto-updated on buffs, then you gain 40s of BL on your DP. Now, let's see what gain you have if dots are updated on cast. Let X denote the time left on your current DP when BL is cast, and let t=0 be (by convention, up to a shift) the time where BL is cast. At time X, you cast a new DP, hasted with BL. The next DP cast is at X+24. If X was less than 16, we have that X+24 <= 40, and this second DP is also hasted by BL. Otherwise, we have X+24 > 40, and the second DP is not hasted. Now, BL is cast at a random time, independently of DP cast, and DP is cast exactly every 24s. This means that the remaining time X on the current DP is a uniform random variable on ]0, 24]. So, we have 2/3 chance of having X less than 16, and 1/3 of having X greater than 16. The mean gain of BL on DP is hence 1/3 * 24 + 2/3 * 48 = 40s of hasted DP. That's identical to the case of auto-update. In other words, when there is no auto-update, the potential gain after the end of the buff is exactly the potential loss before the first recast of the dot. There is no magic value of durations here. This will be always valid, as long as the buff is casted at a random time, independently of the dot casting time. Now, if there is no independence between both, then you could potentially be stucked in a bad synchronization. But with an "equivalent"[*] probability, you could be stucked in a good synchronization. And in fact, you have control of it, and you can ensure in most cases that the synchronization is better that the mean random case. [*]: I don't mean here the probability of being in a good or bad synchronization are equal. I mean that the expected gain, ie. the expectation of the probability of a synchronization times the gain (or loss) of this synchronization, is fundamentally null.
Unless i missed something, this only true if you're casting nothing but DP. The point he was trying to make was, if you cast other things it would be a dps gain if you just reapplied at the end of the dot.

11/10/10, 3:35 PM   #141
Bristine
Glass Joe

Human Priest

Drenden
 Originally Posted by moowalk Having DoTs which don't auto-update is a large increase to shadowpriest dps. It's certainly not worse, since randomly timed bloodlusts (and other effects) will mean an equal amount of lost ticks at the start of the bloodlust as those gained after bloodlust finishes.
That is just not true. If I DP, then BL hits and it autp-updates I get 40 seconds of BL'd DP. If I DP then BL hits immediately after I either A) waste time reDoTing up the target or B) get up to 16 fewer seconds of BL haste.

Making SP's reapply their DoTs when buffs go up instead of allowing them the same freedom other classes get to let buffs passively effect our rotation is unnecessary and an arbitrary nerf. If we take advantage of it, there might be a small DPS increase at the cost of far more micromanagement than any other class. I don't see the point of that, we should get an equal buff from heroism and every other buff as every other class.

Some have argued that we can abuse this: Get a haste trinket and pop it at the beginning of the fight as you put up SWP and you have a hasted SWP all fight if you don't let it fall off. That's ridiculous, and I can't imagine that it is "working as intended" just as our terrible mastery stat wasn't "working as intended" when nobody wanted to put any points into it.

 11/10/10, 6:10 PM #142 ildon Collateral Damage     Nerodius Undead Priest   Whisperwind SW:P haste rolling is clearly not intended, and your understanding of the impact of a randomly timed buff on dot damage isn't correct, either. Because dots don't auto-update, it means you can carry the damage on past the end of the buff, which is not something a nuke can do. I specifically recast VT/DP when Bloodlust had 1 second left to take advantage of this fact. This means that for something like a 15s duration, 45s ICD trinket proc, over the course of the fight the amount of damage lost/gained is going to average out to be basically zero (dependent on the length of the buff relative to the length of your dots). So with a 15s buff and VT, you'd always get exactly one VT out of it if you never overwrote your buffs, it's just that the VT increased damage time relative to the buff duration would be shifted depending on when you cast VT. But because you can watch buffs and consciously recast dots, you could overwrite the VT with 1s remaining on the buff and get between 16 and 29 seconds of buff time on VT. Now, to decide if this is actually worth it you'd have to do a cost:benefit analysis on the loss of the GCD vs. the amount of damage gained, and come up with a rule like "if there's 6s or less left on VT, overwrite it if this buff is up". Generally speaking, I'm pretty sure if you were to just assume that trinket procs and such were "random" and just continue casting normally, over a long time dots not auto-updating is either neutral or only a very small DPS loss (because sometimes, given random input, you'll get increased usage out of a buff equal to the length of the dot, and sometimes you'll get decreased usage equal to the length of the dot, but on average you'll get a change equal to +/- modulus of dot length vs. buff length due to unequal durations) but with intelligent play, and purposefully taking advantage of buffs lining up (such as during blood lust, waiting until you get a trinket proc and then hitting all your cooldowns and refreshing dots at that time, then overwriting them again as Bloodlust reaches 1s remaining) then it is unequivocally a DPS gain to have dots not auto-update, and carry over the damage from when they were initially cast. Even if you couldn't take advantage of dots not auto-updating to increase your DPS, you can't call something a "nerf" when the game never functioned the way you wish it functioned (ignoring ZHC/ToEP stacking days), and was never intended to. When Ghostcrawler originally posted about "auto updating dot stats", he was pretty much only referring to SW:P refresh through Mind flay and Corruption auto update from . SW:P/Corruption haste rolling is almost certainly not intended, but even so, it's unquestionably a DPS gain to anyone who takes advantage of it, so again, not sure how you can call the current situation with dots a "nerf" in any sense.
11/10/10, 9:04 PM   #143
Bristine
Glass Joe

Human Priest

Drenden
 Originally Posted by ildon SW:P/Corruption haste rolling is almost certainly not intended, but even so, it's unquestionably a DPS gain to anyone who takes advantage of it, so again, not sure how you can call the current situation with dots a "nerf" in any sense.
I wasn't clear before, sorry about that. SWP haste rolling isn't a nerf, the other DoT's not auto-updating are a nerf though, or at least I would classify them as such.

To optimize a SP we need to ensure we ce-up DoT's at the end of a buff, meaning a new VT and a new DP at the end of each, assuming of course that this DPS is greater than that of casting whatever else we'd be casting, and as you said, overall it will most likely end up as a minor DPS loss if we ignore the phenomenon. Why I would consider this a nerf is that an SP has an improportionate amount of buff tracking to do as well as re-casting DoTs out of the normal rotation for a minor DPS gain. Lots more work for at most a relatively minor gain is a nerf in my book, especially when the alternative is a relatively minor DPS gain while not having to track our group buffs, you know, like every other class does.

Can we work the mechanic for our advantage? Maybe. If we don't is it a DPS loss? Probably.

 11/11/10, 2:08 AM #144 moowalk Don Flamenco   Moowalk Troll Priest   Khaz'goroth Ildon, can you elaborate on why you think it's a minor dps loss to have dots not updating? It doesn't seem to follow from the rest of your post. Or are you saying it might turn out to be minor dps loss, but is equally likely to be a minor dps gain, and its expected gain/loss is zero? Assume a very long fight, and randomly timed bloodlust. Bristine, if you totally ignore variable buffs and cast dots normally then your dps will be the same as a 'nuker'. If you micro-manage your buffs then a skilled player can produce more dps than a less skilled one. That seems like a good thing to me. I understand how it could be annoying, but it seems a bit of a stretch to call it a nerf.
 11/11/10, 8:39 AM #145 ildon Collateral Damage     Nerodius Undead Priest   Whisperwind I was mostly thinking of Devouring Plague. With randomly timed buffs, it will occasionally miss any buffs shorter than its duration completely, which is most procs and clicky trinkets (with them generally being 15-20 seconds). Depending on the length of the buff, this could come out to a net negative in an infinite length fight. However, the fact that we can recast dots, and the fact that DP has not-insignificant initial damage on cast and has a long duration, means that it benefits the most from intelligently recasting it when you gain a short duration buff, especially now that you can no longer lose a tick for doing so. For the non-intelligent casting method, I calculated the average buff time lost per DP over an infinite length fight to be 3.375s for a 15s buff. For a 20s duration buff, I got 2/3's of a second. You can see it's not really even that significant. Last edited by ildon : 11/11/10 at 8:52 AM.
11/11/10, 10:42 AM   #146
Elimbras
Don Flamenco

Dwarf Priest

Eitrigg (EU)
 Originally Posted by Bristine That is just not true. If I DP, then BL hits and it autp-updates I get 40 seconds of BL'd DP. If I DP then BL hits immediately after I either A) waste time reDoTing up the target or B) get up to 16 fewer seconds of BL haste. [...] I don't see the point of that, we should get an equal buff from heroism and every other buff as every other class.
Once again, your resonning is not valid. Reread my previous post if you want a proof, but the current mechanism is as follows:
1/ You have a potential loss at the beginning of a buff (BL or any other) that you correctly identified: the buff is not applied immediately to your existing dots; you didn't take that one into account.
2/ You have a potential gain at the end of a buff : the buff will still be applied to your existing dots when the buff ends;
3/ The second case happen naturally even if you don't provoke it: your classical rotation / dots refresh will make it happen.
4/ If the buff happen at some random time, and you just follow your classical rotation, the potential loss and the potential gain are exactly equal.

 Originally Posted by Bristine Can we work the mechanic for our advantage? Maybe. If we don't is it a DPS loss? Probably.
In fact, you should correct it to :
Can we work the mechanic for our advantage ? Yes
If we don't, is it a DPS loss ? No[*]
Are we balanced around this clever use of mechanism ? Probably

* : I mean here that we get the exact same benefit of BL as any class which would be spamming a single nuke. Now, it is clear that not working the mechanism to our advantage is a dps loss compared to doing it. But we can't say that we benefit less of heroism than other classes.

 Originally Posted by ildon I was mostly thinking of Devouring Plague. With randomly timed buffs, it will occasionally miss any buffs shorter than its duration completely, which is most procs and clicky trinkets (with them generally being 15-20 seconds). Depending on the length of the buff, this could come out to a net negative in an infinite length fight. However, the fact that we can recast dots, and the fact that DP has not-insignificant initial damage on cast and has a long duration, means that it benefits the most from intelligently recasting it when you gain a short duration buff, especially now that you can no longer lose a tick for doing so. For the non-intelligent casting method, I calculated the average buff time lost per DP over an infinite length fight to be 3.375s for a 15s buff. For a 20s duration buff, I got 2/3's of a second. You can see it's not really even that significant.
I would like to see your computations. This is both against my intuition and any computation I made. Yes, you sometimes won't benefit of the (short) buff. But you will sometimes benefit from it for a longer time than its duration, and the both equalize each other.

11/11/10, 11:05 AM   #147
ildon
Collateral Damage

Whisperwind
 Originally Posted by Elimbras I would like to see your computations. This is both against my intuition and any computation I made. Yes, you sometimes won't benefit of the (short) buff. But you will sometimes benefit from it for a longer time than its duration, and the both equalize each other.
Given a 15s buff duration and a random proc chance, 15/24 times DP will be cast while the buff is up, and gain 9s of additional damage time. 9/24 times DP will be cast while the buff is not up, and "lose" 24s of buff time.

(9*15 -24*9) / 24 = -3.375. This is a comparison vs. a system where dots auto-updated. On Bloodlust, because its duration is longer than DP, obviously this isn't an issue and the gain/loss should net 0.

Edit: Moowalk is correct below. It does net out to 0 gain/loss.

Last edited by ildon : 11/11/10 at 2:34 PM.

11/11/10, 12:42 PM   #148
Bristine
Glass Joe

Human Priest

Drenden
 Originally Posted by moowalk Ildon, can you elaborate on why you think it's a minor dps loss to have dots not updating? It doesn't seem to follow from the rest of your post. Or are you saying it might turn out to be minor dps loss, but is equally likely to be a minor dps gain, and its expected gain/loss is zero? Assume a very long fight, and randomly timed bloodlust. Bristine, if you totally ignore variable buffs and cast dots normally then your dps will be the same as a 'nuker'. If you micro-manage your buffs then a skilled player can produce more dps than a less skilled one. That seems like a good thing to me. I understand how it could be annoying, but it seems a bit of a stretch to call it a nerf.
No, if I continue to not micro manage my DoTs then it won't be the same. BL is 40 seconds, DP ticks for 24 seconds, if I miss the first 23 seconds of BL then I only get 24 seconds of it on the second DP. That is of course worst case scenario, but it is perfectly feasible. Then, even if I do re-up for haste there's no guarantee I get more ticks to make up for the DPS loss of not casting MB while letting DoTs tick at non-buffed rate.

Elimbras can't seem to count to forty, since you do NOT get that buffed DoT time back automatically. It isn't "I lose 16 seconds at the beginning of the buff and get it back at the end" I lose 16 seconds at the beginning and get a grand total of 24 hasted DoT seconds unless I re-apply at the end losing potential damage from MF.

I'll try to make this as clear as possible: I DP then BL hits. I re DP as usual and get 24 seconds of buffed time, as when I would reapply DP again as usual BL would be over. I don't get that time back. At best I can reapply DP when BL starts and then normally during BL, then again right before it fades giving me a potential 72 seconds of buffed DP at the cost of the three GCD's I would spend on my normal rotation.

At best, it could be a minor DPS boost, but most likely it wont be without intense micro, and even then compared to any class that gets the buffs applied automatically, it is way more to keep track of. Someone needs to do the math on time lost reapplying buffs and that DPS loss compared to the DPs gain of the buffs being applied, I would wager that it is a very minor bonus and that we would be far better off overall to just have buffs apply automatically.

That is also before VT. If we apply both that's two GCD's that would otherwise be spent on stacking DPs with DoT's and MF/SWD/MB lost. It is a definite loss overall if we don't, and only a minor gain if we do. That's a nerf in my book.

Last edited by Bristine : 11/11/10 at 12:57 PM.

11/11/10, 1:50 PM   #149
moowalk
Don Flamenco

Troll Priest

Khaz'goroth
 Originally Posted by ildon Given a 15s buff duration and a random proc chance, 15/24 times DP will be cast while the buff is up, and gain 9s of additional damage time. 9/24 times DP will be cast while the buff is not up, and "lose" 24s of buff time. (9*15 -24*9) / 24 = -3.375. This is a comparison vs. a system where dots auto-updated. On Bloodlust, because its duration is longer than DP, obviously this isn't an issue and the gain/loss should net 0.
The mistake you've made there is that you don't "lose 24s of buff time", you lose 15 seconds of buff time since the buff is only 15 seconds.

Your final calculation becomes (9*15 -15*9) / 24 = 0

 I'll try to make this as clear as possible: I DP then BL hits. I re DP as usual and get 24 seconds of buffed time, as when I would reapply DP again as usual BL would be over. I don't get that time back. At best I can reapply DP when BL starts and then normally during BL, then again right before it fades giving me a potential 72 seconds of buffed DP at the cost of the three GCD's I would spend on my normal rotation.
You seem to be arguing something else here, which is that bloodlust length is not a multiple of devouring plague length which leads to annoyingly timed refreshes. This could definitely make optimizing dot refreshes difficult, but it won't lead to a worse outcome, it will just make it harder to get a positive outcome.

11/11/10, 5:21 PM   #150
ilkori
Von Kaiser

Blood Elf Priest

Silvermoon
 Originally Posted by moowalk The mistake you've made there is that you don't "lose 24s of buff time", you lose 15 seconds of buff time since the buff is only 15 seconds. Your final calculation becomes (9*15 -15*9) / 24 = 0
So what you are saying is that the 24*9 does not compare to a system where haste auto-updates. Apples to apples and all that.

So yes, there is a net zero effect on average if we do not manipulate our cast sequence to take advantage of the effects.

"It's kind of fun to do the impossible." - Walt Disney

 Elitist Jerks Shadowpriest FAQ for 4.0