PVE Raiding Compendium - Page 123

Trickykid · 05/23/08, 4:10 PM

Originally Posted by Torq

I would say the movement effect is more drastic on haste, if you're looking at it from a per-cast standpoint. Unless every bolt is getting interrupted (hence leaving you with 0 dps anyway), crit and dmg have a greater effect per-bolt (or sequence of bolts) than haste does, under less-than-ideal circumstances. Going back to the breakpoint discussion, unless you can squeeze in an extra bolt every cast chain over what you'd get without haste, crit or dmg are going to give you greater gains.

The problem with haste is that the benefit can only be quantified with integer values. Going from 15 bolts to 15.99 bolts before having to move is 0 benefit, since 99% of a bolt is not 1 bolt. However, going from 5 crits to 6 in that same timeframe (on average) is a benefit, even though it is subject to the RNG effect. Going from 3000 average damage to 3050 average damage for each of those 15 bolts is even better, because it is even less subject to the RNG effect.

To be honest, unless it's a straight burn fight (aka Brutallus, unless you get extremely unlucky with Burns), dmg > haste > crit. On a percentage increase basis, it's actually dmg > crit > haste, but since crit is so expensive on items, it drops in value.

It might be helpful to generate some sort of metric for fights: Average # of movements and average fight length, to give you an average burn length (length / movements) to scale the value of haste accordingly. It'd at least give you a slightly more realistic idea of whether or not haste will give you an actual benefit (integral increase in bolts cast).

I mean, if you take a fight like IC, and see that the length is on the order of 10 min (very roughly), you might think 10% haste will give you a 10% dps increase, or an extra 24 bolts. However, if you have to move ~20 times, on average, then your average burn period is ~30 sec. A 30 second burn period with 10% haste will net 1 extra bolt over 0 haste, meaning haste does give you a benefit. However, due to movement, you're only squeezing in an extra 20 bolts, a net 4-bolt loss over the expected gain. The scaled gain would be 20/24 or 83% of the expected gain.

But see, there's a fallacy of logic. Movement will always interrupt burn periods, regardless of the amount of haste, and with roughly the same probability. In other words, at any given point in time, you are casting, meaning a movement is interrupting a cast. So, your actual dps will always be lower than the expected. Again, like I stated above, the problem with haste is it operates on gains in the bounds of integers. Unless you're always fitting an extra cast in your burn periods, you're losing the full benefit of haste. It's just less drastic when you are getting an extra bolt than when you don't, because 0 extra bolts means every point of haste has gone to waste, whereas x extra bolts just means some fraction of the haste has gone to waste..

Hasted or not, that .5 sec of a shadowbolt is lost; and in fact, if it's hasted, .5 sec of casting is worth more than .5 sec of unhasted casting; you would agree this is the case, especially when under the influence of heroism, etc.

The point I'm making is that the breakpoints go both ways. If at time T you have to stop casting, that can either fall on the side of benefiting or costing a hasted lock (in other words the time window cuts more or less into either a hasted or unhasted sequence of casts). It doesn't make sense to just consider the times when it falls on the negative side.

I did up a very simplified table to show this. This value changes per T, of course, but I broke it down into 6 windows. So 7 to 13 means if the window has an equal chance of being any T between 7 and 13 seconds. Time T you are forced to stop casting and the actual % more casts from the hasted sequence:

Time slice      Percent more SBs with 10% haste
7 to 13	        10.7%
12 to 18        11.2%
17 to 23        9.6%
22 to 28        10.5%
27 to 33        9.9%
32 to 38        9.6%
37 to 43        10.2%
42 to 48        9.8%
47 to 53        10.2%
52 to 58        9.9%

As you can see it goes above and below 10% depending on the window. What do we learn? The shorted the window the larger the variation as it converges to the expected 10%. There is no trend to stay below 10%, however, which means you are just as likely to see an OVER-valued haste than under-valued.

Anexus · 05/23/08, 4:13 PM

I think we all agree that you do not get a 100% benefit of haste. But you do not get a 100% benefit of crit and spelldmg either. What we need is some relative value that shows how much you benefit from each stat, on avarage. How big is the difference? Do we gem 12spelldmg in everything or do we put 2x(6spelldmg+5haste) and get the socket bonus of 4 spelldmg? I'll leave the math to somebody who knows what he's doing.

tetracycloide · 05/23/08, 4:19 PM

Originally Posted by Torq

Hasted or not, that .5 sec of a shadowbolt is lost; and in fact, if it's hasted, .5 sec of casting is worth more than .5 sec of unhasted casting; you would agree this is the case, especially when under the influence of heroism, etc.

You are contradicting yourself here. Your argument is that in a movement based fight that haste looses significantly more value than other DPS stats because the loss in cast time is magnified by the fact that it is lost hasted cast time. However this ignores the opposite effect where haste increases the value of the avaiable cast time. There is no logical reason to believe that one effect would be greater than the other if haste makes all of your time equally more valuable.

The actual reason anyone would agree that loosing heroism time to movement is bad is because it is not a static buff that applies for the entire fight. Loosing heroism time is bad because heroism is a temporary buff on a cooldown. The comparison of heroism time to a static buff like spell haste is not apples to apples.

Torq · 05/23/08, 4:57 PM

Originally Posted by Trickykid

The point I'm making is that the breakpoints go both ways. If at time T you have to stop casting, that can either fall on the side of benefiting or costing a hasted lock (in other words the time window cuts more or less into either a hasted or unhasted sequence of casts). It doesn't make sense to just consider the times when it falls on the negative side.

I did up a very simplified table to show this. This value changes per T, of course, but I broke it down into 6 windows. So 7 to 13 means if the window has an equal chance of being any T between 7 and 13 seconds. Time T you are forced to stop casting and the actual % more casts from the hasted sequence:

As you can see it goes above and below 10% depending on the window. What do we learn? The shorted the window the larger the variation as it converges to the expected 10%. There is no trend to stay below 10%, however, which means you are just as likely to see an OVER-valued haste than under-valued.

But you're missing the point here. The "expected value" is the "spreadsheet value" which assume no movement at all. We're looking at it from two different relative standpoints. I'm looking at it relative to the spreadsheet expected value, you're looking at it relative to the fight's expected value. One assumes we never lose time to movement, the other assumes we always lose time to movement.

And no, the breakpoints never go both ways. You either have the same number of hasted casts as unhasted, or you have more hasted casts than unhasted. You can never have fewer hasted casts than unhasted for the same window of time. So, given a window for casting, you either gain part of the haste benefit, or none at all; as in you either cast at least one more bolt in that window, or you had to move and cast the exact same number as you would have without haste.

Again, I'm not saying haste is bad, just that it's no where near as valuable as the spreadsheet weighs it to be. If it's no where near as valuable as the spreadsheet weighs it to be, then point for point on items (aka relative to item budgets), +dmg is worth more.

Originally Posted by tetracycloide

You are contradicting yourself here. Your argument is that in a movement based fight that haste looses significantly more value than other DPS stats because the loss in cast time is magnified by the fact that it is lost hasted cast time. However this ignores the opposite effect where haste increases the value of the avaiable cast time. There is no logical reason to believe that one effect would be greater than the other if haste makes all of your time equally more valuable.

The actual reason anyone would agree that loosing heroism time to movement is bad is because it is not a static buff that applies for the entire fight. Loosing heroism time is bad because heroism is a temporary buff on a cooldown. The comparison of heroism time to a static buff like spell haste is not apples to apples.

I don't see how this is a contradiction. I'm merely emphasizing that haste is not worth as much as it's "expected" to, relative to the spreadsheet values and +dmg equivalences. And the problem is that haste doesn't make all your time more valuable. Lost time is always lost time, regardless of whether or not it's a fraction of a cast or a whole cast. Casting time is only worth the number of wholte casts you can fight in there, and any excess cast is wasted.

Trickykid · 05/23/08, 6:04 PM

Originally Posted by Torq

But you're missing the point here. The "expected value" is the "spreadsheet value" which assume no movement at all. We're looking at it from two different relative standpoints. I'm looking at it relative to the spreadsheet expected value, you're looking at it relative to the fight's expected value. One assumes we never lose time to movement, the other assumes we always lose time to movement.

And no, the breakpoints never go both ways. You either have the same number of hasted casts as unhasted, or you have more hasted casts than unhasted. You can never have fewer hasted casts than unhasted for the same window of time. So, given a window for casting, you either gain part of the haste benefit, or none at all; as in you either cast at least one more bolt in that window, or you had to move and cast the exact same number as you would have without haste.

Again, I'm not saying haste is bad, just that it's no where near as valuable as the spreadsheet weighs it to be. If it's no where near as valuable as the spreadsheet weighs it to be, then point for point on items (aka relative to item budgets), +dmg is worth more.

You're misunderstanding my use of "both ways". You are focusing only on the possibilities that lead to less than 100% of the expected benefit. It goes both ways from 100% though, and if you look at my post in more detail, you can see that there are cut offs with 10% haste where you get more than 10% more bolts, which means there are as many possibilities of it being overvalued as undervalued.

However, you're claiming that I'm comparing this to an unhasted lock, not to the spreadsheet numbers. This doesn't matter given what we just saw above. Based on those numbers, haste gets very very close to (if not, in some situations more than) the expected gain. By this I mean if we have 10% haste, we get roughly 10% increased bolts (refer to my previous post on the variations in this value), which is what the spreadsheet is assuming as well (roughly). Therefore, spell damage and crit don't get any significant (or arguably any) relative advantage here, and certainly not to the point of dropping its worth by the amount you are claiming.

Torq · 05/23/08, 6:19 PM

Originally Posted by Trickykid

You're misunderstanding my use of "both ways". You are focusing only on the possibilities that lead to less than 100% of the expected benefit. It goes both ways from 100% though, and if you look at my post in more detail, you can see that there are cut offs with 10% haste where you get more than 10% more bolts, which means there are as many possibilities of it being overvalued as undervalued.

However, you're claiming that I'm comparing this to an unhasted lock, not to the spreadsheet numbers. This doesn't matter given what we just saw above. Based on those numbers, haste gets very very close to (if not, in some situations more than) the expected gain. By this I mean if we have 10% haste, we get roughly 10% increased bolts (refer to my previous post on the variations in this value), which is what the spreadsheet is assuming as well (roughly). Therefore, spell damage and crit don't get any significant (or arguably any) relative advantage here, and certainly not to the point of dropping its worth by the amount you are claiming.

No, the spreadsheet assumes we get a 10% increase to shadowbolt dps, as that's the theoretical gain of haste (10% haste = 10% more dps). I'm saying that the spreadsheet is assuming 100% dps time, meaning you never move. From that, I'm claiming that realistically, you always have to move some time during a fight. Therefore, no matter what, your dps will always be lower than the spreadsheet value, because you had to move.

You're claiming that by moving, you're somehow doing more dps than the spreadsheet claims. This would imply that you're casting for >100% of the fight, which is impossible.

The spreadsheet is the absolute upper bound (assuming the RNG behaves on crits, etc). You cannot achieve greater dps than this.

So you may get close to 10% more dps than an unhasted lock who has to move the exact same as you do during a fight; but you're not getting more than the spreadsheet value for the haste. This implies that since the "accepted" value of haste is based on what the spreadsheet tells you (based on your gear level) that haste is, in fact, less valuable than the spreadsheet claims, purely because you always have to move.

And, if you're hovering on that 1.15:1 haste:dmg value ratio, your effective haste value ratio may well be somewhere around 1:1, meaning you're better off gemming for pure damage than pure haste, because haste costs more as an item stat.

Trickykid · 05/23/08, 6:28 PM

Originally Posted by Torq

No, the spreadsheet assumes we get a 10% increase to shadowbolt dps, as that's the theoretical gain of haste (10% haste = 10% more dps). I'm saying that the spreadsheet is assuming 100% dps time, meaning you never move. From that, I'm claiming that realistically, you always have to move some time during a fight. Therefore, no matter what, your dps will always be lower than the spreadsheet value, because you had to move.

You're claiming that by moving, you're somehow doing more dps than the spreadsheet claims. This would imply that you're casting for >100% of the right, which is impossible.

The spreadsheet is the absolute upper bound (assuming the RNG behaves on crits, etc). You cannot achieve greater dps than this.

So you may get close to 10% more dps than an unhasted lock who has to move the exact same as you do during a fight; but you're not getting more than the spreadsheet value for the haste. This implies that since the "accepted" value of haste is based on what the spreadsheet tells you (based on your gear level) that haste is, in fact, less valuable than the spreadsheet claims, purely because you always have to move.

And, if you're hovering on that 1.15:1 haste:dmg value ratio, your effective haste value ratio may well be somewhere around 1:1, meaning you're better off gemming for pure damage than pure haste, because haste costs more as an item stat.

The fact that the spreadsheet shows you the upper bound has nothing to do with the relative weight of the the stats that make up that damage. That was what I was trying to show you in previous posts.

The relative weights from the sheet can be shown accurate at other DPS-times (at least for simple SB spam) if we see those stats contributing in similar fashion at other levels. All I have to do (and did, in previous posts) is show that haste contributes the same % increase in fights with arbitrarily shortened time windows as it does with 100% dps time.

I think I have shown in a straightforward way that even with shortened time windows in the worst case scenarios, haste will show very close to 10% increase in bolts (and at best, better than 10% increase). Given that +dmg and +crit add the same % increase at 100% and less dps-times, they can all be compared relatively for any movement fights.

edit: Ok, one more post...

Torq · 05/23/08, 6:43 PM

Originally Posted by Trickykid

The fact that the spreadsheet shows you the upper bound has nothing to do with the relative weight of the the stats that make up that damage. That was what I was trying to show you in previous posts.

I was under the impression that the stat weights are calculated on the absolute values of your damage, rather than relative values. And also, I'm still pretty firmly convinced that relative to an unhasted caster, the hasted caster always either gets the same amount of casts or more; not strictly always more.

In other words, if you stuck a caster with x damage and y% haste next to a caster with x damage and 0 haste, the caster with haste would come out on top, but not by y%. The application would be to take caster A with x damage and y% haste and compare him to caster B with x+(y% as dmg) dmg and 0% haste.

I think I'll run a little sim of this and see what happens. My expectation is that B would come out on top, pretty consistently, given any amount of random periodic movement.

[E] Okay, results of the sim:

00:00:00.0000>> Started
00:00:00.0000>> 0 (hasted)-> 2.2828604516503,1600
00:00:00.0000>> 1 (damage)-> 2.5,1736.3636363636
00:30:01.0595>> Finished - Duration 1801.5955629357 seconds
00:30:01.0595>> 0 (hasted)-> 926400
00:30:01.0595>> 1 (damage)-> 1111272.7272727
00:30:01.0595>> Ratio:  0.8336387434555

Caster 0 has 150 haste and 1600 damage, Caster 1 has 0 haste and 1736... damage (assumes 1 haste is worth 1.1 damage, or .90~ haste is 1 damage). Movement comes randomly every 20-30 seconds, for 2-4 seconds per move. The final ratio is total Caster 0 damage / total Caster 1 damage

Second iteration, with same dmg and haste stats, but much longer burn periods (60-90 sec)

00:00:00.0000>> Started
00:00:00.0000>> 0 (hasted)-> 2.2828604516503,1600
00:00:00.0000>> 1 (damage)-> 2.5,1736.3636363636
00:30:00.0189>> Finished - Duration 1800.1893230423 seconds
00:30:00.0189>> 0 (hasted)-> 1067200
00:30:00.0189>> 1 (damage)-> 1272754.5454545
00:30:00.0189>> Ratio:  0.83849631793605

Still about the same ratio. So, unless I'm doing something wrong, it appears that +dmg comes out on top every time.

[E] Forgot to include number of casts in there. 0 = 667 and 1 = 733; which doesn't really make any sense at all. I'll have to look over my script again, or rewrite it (wrote it at work and forgot to bring it back with me) and figure out what's wrong.

Trickykid · 05/23/08, 7:08 PM

Originally Posted by Torq

I was under the impression that the stat weights are calculated on the absolute values of your damage, rather than relative values. And also, I'm still pretty firmly convinced that relative to an unhasted caster, the hasted caster always either gets the same amount of casts or more; not strictly always more.

In other words, if you stuck a caster with x damage and y% haste next to a caster with x damage and 0 haste, the caster with haste would come out on top, but not by y%. The application would be to take caster A with x damage and y% haste and compare him to caster B with x+(y% as dmg) dmg and 0% haste.

I think I'll run a little sim of this and see what happens. My expectation is that B would come out on top, pretty consistently, given any amount of random periodic movement.

Finally attempt to clarify what I'm saying:

Two simple scenarios with two locks playing optimally, maximizing dps time, etc.

Actors:
-) Lock1 with 0% haste
-) Lock2 with 10% haste.

Scenario 1: Stand and nuke fight.
Results:
-) Lock1 has X casts
-) Lock2 has 1.1*X casts

10% more casts from 10% haste.

Scenario 2: Movement fight with slighly varying windows of DPS time.
Results:
-) Lock1 hax Y casts
-) Lock2 has between 1.09*Y and 1.11*Y casts (rough estimation from the numbers I posted in the lock thread).

9-11% more casts from 10% haste.

Let's go through the reasoning for this variation. Some of the time windows, depending upon their exact length, might leave Lock2 with the same number of casts as Lock1 (simple example is 3 seconds where each can only fit in 1 bolt). There are other time windows when Lock 2 gets multiple more bolts in (simple example is 7 second fight, where Lock1 has 2 bolts but Lock2 gets 3 bolts in). In the first example, haste yields 0% more casts, but in the second it yields 50% more casts. In that one specific time window, 10% haste allowed for 50% more damage. Due to this variation, where haste will give you more or less than the expected return (in the case of 10% haste, less or more than 10% increased casts), haste COULD add more damage than expected.

Now, to move on to the relative weights of these stats. The spreadsheet adds up the gain from each stat with 100% dps time and then compares them relatively (because no one cares how much DPS you gain from 1 spell damage, they care how much crit and haste are worth compared to spell damage). For this reason, it doesn't really matter that the sheet only looks at 100% stand and nuke fights as much, so long as movement fights value each stat relatively the same. I find it pretty clear (from both intuition and napkin math) that haste yields (on the average) as much in movement fights as in stand and nuke fights. Given this, the relative weights from the sheet hold true for movement fights as well and do not strictly suffer from arbitrary time slices.

PSGarak · 05/23/08, 7:23 PM

Originally Posted by Torq

You're claiming that by moving, you're somehow doing more dps than the spreadsheet claims. This would imply that you're casting for >100% of the fight, which is impossible.

No, what he's claiming is that under conditions that cause you perform less than the spreadsheet, the breakpoint will allow haste to net you an extra shadowbolt as often as it doesn't, meaning under some windows 10% haste is more than 10% damage. This is not higher than the spreadsheet value if the time window also includes the run time, it's just closer to the spreadsheet value than an unhasted warlock. You're misunderstanding the manner in which he is trying to contradict you. You are saying that a hasted warlock will always drop by more than an unhasted one during movement, which is incorrect.

If you take a shadowbolt from 2.5 to 2.4, and a movement breakpoint occurs on a multiple of 2.5s that is not a multiple of 2.4s, then you stop a hasted shadowbolt early and negate the effect of haste. On the other hand, if the breakpoint occurs on a multiple of 2.4 that is not a multiple of 2.5, the haste will allow you to squeeze off an extra shadowbolt that you otherwise could not have, gaining you one whole cast in a time window smaller than that required to get an extra cast under normal situations.

Furthermore, haste nets you an extra cast more often than it is wasted because, by stint of being a shorter time interval, there are more multiples of 2.4s than 2.5s in an interval. The difference is probably incredibly small (second-order on the difference?), but the measure of time spent after a 2.4s cast before a 2.5s cast, where 2.4s gains an extra bolt, is larger than the measure of time in the opposite situation because 2.4s casts are more frequent... I think. It might be the same, since the situations could be viewed as symmetric (one before, one after). Anyways, the point is, movement does not affect haste more than any other DPS stat.

Torq · 05/23/08, 9:30 PM

Originally Posted by PSGarak

Furthermore, haste nets you an extra cast more often than it is wasted because, by stint of being a shorter time interval, there are more multiples of 2.4s than 2.5s in an interval. The difference is probably incredibly small (second-order on the difference?), but the measure of time spent after a 2.4s cast before a 2.5s cast, where 2.4s gains an extra bolt, is larger than the measure of time in the opposite situation because 2.4s casts are more frequent... I think. It might be the same, since the situations could be viewed as symmetric (one before, one after). Anyways, the point is, movement does not affect haste more than any other DPS stat.

See, I can agree with this, and see this as being the case. Having casts as smaller intervals would, I suppose, allow for them to be "less likely" to be interrupted. That part of logic makes sense.

However, I wouldn't say movement doesn't affect haste more than any other stat. Movement doesn't affect crit or damage at all, considering neither of them makes a change in the number of spells landed in a given time period. In other words, whether or not you added 10% to crit, you still were going to land those same x spells anyway; just an additional 10% of them are now crits. I guess the same could be said for haste, in an overall fashion, if not in a specific fashion.

Anexus · 05/23/08, 9:42 PM

Originally Posted by Torq

00:00:00.0000>> Started
00:00:00.0000>> 0 (hasted)-> 2.2828604516503,1600
00:00:00.0000>> 1 (damage)-> 2.5,1736.3636363636
00:30:01.0595>> Finished - Duration 1801.5955629357 seconds
00:30:01.0595>> 0 (hasted)-> 926400
00:30:01.0595>> 1 (damage)-> 1111272.7272727
00:30:01.0595>> Ratio:  0.8336387434555

00:00:00.0000>> Started
00:00:00.0000>> 0 (hasted)-> 2.2828604516503,1600
00:00:00.0000>> 1 (damage)-> 2.5,1736.3636363636
00:30:00.0189>> Finished - Duration 1800.1893230423 seconds
00:30:00.0189>> 0 (hasted)-> 1067200
00:30:00.0189>> 1 (damage)-> 1272754.5454545
00:30:00.0189>> Ratio:  0.83849631793605

would you mind adding a calucation for a char with x haste and y extra spelldmg? just to see how the hybrid comes out. when you get this thing fixed ofc

Torq · 05/23/08, 10:40 PM

Here's something similar, and I believe I fixed whatever the bug was. Just rewrote the whole thing from scratch:

Here's a 30-min sim with 4 toons (+1 baseline) and haste being considered worth 1.15 points of dmg:

000:00:00.000 >> Start:  4 characters, even distribution from 150 haste to 0 and 1600 to 1772.5
000:00:00.000 >> 0 >> Cast: 2.500	Damage: 1772.500
000:00:00.000 >> 1 >> Cast: 2.423	Damage: 1715.000
000:00:00.000 >> 2 >> Cast: 2.351	Damage: 1657.500
000:00:00.000 >> 3 >> Cast: 2.283	Damage: 1600.000
000:00:00.000 >> Baseline >> Cast: 2.500	Damage: 1600.000
000:30:00.358 >> Finished
000:30:00.358 >> 0 -> Cast: 2.500	Damage: 1772.500
000:30:00.358 >> 0 -> Casts: 683	Damage: 1210617.5
000:30:00.358 >> 0 -> Percentage of baseline: 110.781%
000:30:00.358 >> 1 -> Cast: 2.423	Damage: 1715.000
000:30:00.358 >> 1 -> Casts: 700	Damage: 1200500
000:30:00.358 >> 1 -> Percentage of baseline: 109.855%
000:30:00.358 >> 2 -> Cast: 2.351	Damage: 1657.500
000:30:00.358 >> 2 -> Casts: 721	Damage: 1195057.5
000:30:00.358 >> 2 -> Percentage of baseline: 109.357%
000:30:00.358 >> 3 -> Cast: 2.283	Damage: 1600.000
000:30:00.358 >> 3 -> Casts: 742	Damage: 1187200
000:30:00.358 >> 3 -> Percentage of baseline: 108.638%
000:30:00.358 >> Baseline >> Cast: 2.500	Damage: 1600.000
000:30:00.358 >> Baseline >> Casts: 683	Damage: 1092800

Daeven · 05/24/08, 12:00 AM

Originally Posted by PSGarak

Furthermore, haste nets you an extra cast more often than it is wasted because, by stint of being a shorter time interval, there are more multiples of 2.4s than 2.5s in an interval.

Anyways, the point is, movement does not affect haste more than any other DPS stat.

The problem with this line of reasoning is that it does not hold true when applied to the specific mechanics of many WoW encounters, and even on ones where it might, trying to accurately model the effect is unreliable. That is, to provide a modelling basis for the statement "under conditions that cause you to perform less than the spreadsheet, the breakpoint will allow haste to net you an extra shadowbolt as often as it does not" is problematic at best, because when there are pervasive (i.e. frequent or raid-wide) chance-based interruptions in a WoW encounter, they are usually sufficient to cause equal negative effect for both hasted and non-hasted casters. Trying to model the relative negative effect in such random encounters based on "multiples" of cast time is too unreliable to be useful.

This is why I emphasised in my original post that the type/extent of interruptions for each fight must be closely understood. It is relatively common in WoW encounters to be given fixed/reliable segments of time to DPS, often with interplay from random variables, leading to different outcomes in relation to haste:

- Fights like Najentus and Kalecgos which are dominated by fixed segments of casting time, in some cases with low probability (e.g. plucking a spine) interruptions. These kinds of interruptions can be safely ignored for modelling purposes and the damage contribution of haste is well understood. For these fights, as I explained in my original post - if the level of haste does not hit a certain value/range suitable for the segments of casting time allowed by the encounter mechanics, you will see any one of minimal (unduly low haste rating), wasted (in the case where a haste rating will provide let's say 1.7 extra casts per segment which in fact is just 1 extra cast), or zero (with insufficient haste rating) contributions to total damage. Obviously the longer the segment the greater the ability to realise gains from realistic/typical levels of haste.

- Fights like Teron Gorefiend, Kaz'rogal and Brutallus which are generally not based on phases, with only low probability/frequency or absolute (e.g. Teron ghosting) interruptions, allowing long periods of casting to occur. These fights are ideal for haste, and will come closest to fully realising the theoretical damage increases.

- Fights that involve a well-defined mix of casting periods versus movement/interruptions, such as Supremus, Gurtogg Bloodboil and Felmyst. The value of haste in these common fights is often limited by the short durations between interruptions, threat mechanics, etc. In the case of Gurtogg Bloodboil for example, the very frequent movement during bloodboils, small window for insigificance casting, and the fact that interruptions do not require an immediate cancellation of casting can lead to a scenario where typical BT/MH levels of haste constitute a zero increase in total damage dealt over having lesser/no haste (in fact its a net loss in damage due to the stats like +dmg/crit given up to gain the haste). This same concept applies to other bosses like Supremus - trying to say that a 2.4 sec hasted Warlock gains any better ability to move from random punches or volcanos and therefore gain/save a cast is invalid.

- Fights that involve periodic interruptions (to varying degrees) due to chance-based mechanics such as Rage Winterchill, Mother Sharaz and Illidari Council. Due the entirely random and ongoing nature of these mechanics, it is very unreliable to assume the "saved/extra" number of casts a Warlock of a given haste rating might have versus another. Note that I am trying to position this thinking above the typical "haste versus no haste" analysis - for example, consider the negligible advantage on Illidari Council a Warlock with 4% haste has over one with 6% haste, and so on. Likewise, a fight like Mother Sharaz is too random with Fatal Attraction to consider in the context of "multiples of cast time". In these cases the effect of haste is effectivley bound to the chance-based nature of the encounter - if the outcome (or risk related to the outcome) is considered poor, then you could argue that for the casts you do successfully get off, they would stand to benefit more from increased +dmg or crit chance that would otherwise be sacrificed for the haste.

- Some fights have such small, fixed segments of uninterrupted casting time that I treat them as an exception. For example, the high frequency of silences by Azgalor can mean either substantial benefit if gaining an extra cast or an absolute zero benefit if not having an adequate haste rating.

In summary, haste is certainly effected by the dynamics of a given WoW raid encounter, significantly more than other stats because it is the only one that does not realise its damage potential on impact. It requires segments of time, however short or long, for the cast volume gain/save to be made and the damage contribution realised - and therefore it is inherently bound to factors that halt/limit casting sequences.

Modelling the effect of haste based on either the fixed or average/probable segments of time that might be achievable in a fight (given its dynamics) is arguably the best context to understand its true impact. Trying to model the effect in terms of relative "multiples of cast time" is highly unreliable - in other words, to say that a Warlock is better off to have 2.4 versus 2.5 sec cast time because it lets them cast two times and avoid a random effect that requires/enforces instant cancellation of casting after 4.8 seconds versus 5 seconds is a poor basis for modelling and does not represent the reality of most WoW raid encounters.

If anything, one can argue that increasing levels of haste provide exponentially higher levels of damage potential due to its compressing effect on segments of time - that is, improving haste to a (modelled) optimum level/range which puts you "within" the widest number of casting segments implemented across the bosses relevant to your guild's raids will lead to maximising the damage potential of haste. Any less or more than the optimum level is in fact damage potential lost, given it could be better spent on other stats. Hopefully this is food for thought anyway, this forum is about advancing both theory and practice, so given the complexity and speciality of haste relative to other stats (i believe Blizzard intentionally designed it this way), the more it is debated the better.

Zed · 05/24/08, 6:23 AM

If fight or casting period lengths are independent from your haste (which they are), you are equally likely to see a greater-than-spreadsheet benefit from haste as a smaller one. The average benefit will be the theoretical value. I suspect you are assuming worst case scenarios here.

Additionally on movement fights, having haste will actually make the chance of you having to cancel a cast smaller, as shown above by PSGarak. Say your cast time is x, and a boss throws a random ability at you that forces you to move in y seconds. Obviously the only interesting case is when y < x otherwise you simply finish your cast, move, then resume casting at your new location. Time spent with movement can be modeled by subtracting it from the fight length which would remain to be independent, the expected benefit of haste staying at the spreadsheet value.
As for cases when y < x the chance that you do not have to cancel a cast before moving, y/x will be greater for a hasted cast time (smaller x), thus you have an extra benefit here from having haste.

The only way I see you could be right is if canceling a hasted cast would be a greater loss than a non-hasted one, which I don't see how could be true.

Daeven · 05/24/08, 7:15 AM

Originally Posted by Zed

If fight or casting period lengths are independent from your haste (which they are), you are equally likely to see a greater-than-spreadsheet benefit from haste as a smaller one. The average benefit will be the theoretical value. I suspect you are assuming worst case scenarios here.

Forgive me if I keep pressing in on a certain angle here, but I am keen to get to the heart of myths about haste that have existed for a long time now - that haste in and of itself at any value, leads to a damage increase.

Case 1: Uninterrupted Casting Windows: As I demonstrated using examples from T6 bosses, if casting windows are independent of haste (which of course is true) you are most certainly not equally likely see a greater than spreadsheet effect as you will a reduced effect. Unless your particular haste rating means you will cast a full extra spell/s within a given uninterrupted period, you will always realise a zero increase in damage for that period.

Case 2: Cancelling Casts: Again, as I made examples of in my post above, there are actually few T6 fights in WoW that require you to instantly cancel a spell and move. And for those that do, the effect requiring the movement is entirely random and unreliable to draw conclusions from re: the benefit of haste. While a very high value of haste would be most likely to lead to some benefit, it still is no guarantee - depending on the intervals of the interruption, no damage benefit can easily be realised. And again, these types of fights are rare.

There is no doubt that haste is a valid stat to build up, and it certainly comes into its own in Sunwell-type gear, because the haste is often coming for "free" or at little opportunity cost. But haste needs to be considered carefully before you will see real returns. My original raising of this issue was based on Warlocks I see donning 2-3 pieces of haste gear and achieving 3-4% haste (~2.4 sec sbolts) because Warlock damage modelling spreadsheets show it as quite a substantial increase in dps. However, particularly at these lower levels, haste is a stat that can quite clearly be demonstrated to lead to very limited (or no) increase in overall damage for many T6 fights in WoW.

Zed · 05/24/08, 8:33 AM

I will try to use your Naj'entus example to demonstrate where you are wrong.

In the first segment, you are ignoring the fact that the non-hasted warlock will also lose one cast (or more) due to applying curses, aggro build and positioning. This can be modeled by modifying the cast period length to 60-x where x is the time spent with those. If the integer quotient of (60-x)/2.4 is greater than that of (60-x)/2.5 you will see a greater than expected benefit from haste, which you completely ignore.

In the fourth segment likewise, it is possible that the hasted warlock will achieve an extra cast and the boss will die while the non-hasted warlock's bolt is still casting or in mid-air. In your example you say the boss dies in 20 seconds (x=40), which is an integer multiple of 2.5 (and not 2.4), thus you are setting up the example in your favor. Counter example: boss dies in 24 seconds. You achieved an extra cast in 24 seconds instead of 60, thus gaining a larger than average benefit from haste.

Sure if you are able to predict cast segment lengths then you could adjust your haste to have maximum effect. But with all the randomness in encounters, lag etc. this is impossible. The infinite time frame modeling used in the spreadsheet actually gives a good estimate for the value of haste.

Kalle · 05/24/08, 8:54 AM

Let's consider a boss that forces you to move at random times. We divide the fight in three parts: normal dpsing, moving and time lost because having to cancel a spell. The chance you have to cancel after x seconds of casting is equal for all x with 0 < x < casttime. So you loose on average 0.5*casttime, which you would otherwise have spent dpsing plus some time required to move. Or in terms of damage: damage = dps*duration_of_fight - time_spent_moving*dps - 0.5*casttime*dps. So your dps with movement will actually be higher if your casttime is shorter (but the difference is very small).

If the time between movements is fixed but large the distribution of x will also be uniform because of varying lag, not smashing buttons at exactly the same time etc. So the same formula applies.

rutiene · 05/24/08, 10:32 AM

Originally Posted by Zed

If fight or casting period lengths are independent from your haste (which they are), you are equally likely to see a greater-than-spreadsheet benefit from haste as a smaller one. The average benefit will be the theoretical value. I suspect you are assuming worst case scenarios here.\

Assuming that you mean that the theoretical value is that of which is given to us by the spreadsheet, that would have to mean that the theoretical value for haste includes the probability for the haste to be 'useful'. Which is entirely too dynamic and specific to an encounter to be modeled properly. Thus, the spreadsheet truly only gives us the best benefit, and not the average. Since haste is of the nature of being either beneficial or not, in that there's no happy medium it can fall under, this best benefit theoretical value does not give anywhere close to the whole picture and is not comparable with the other theoretical values. Thus it is not wise to judge the value of haste on your DPS based on theoretical values from the spreadsheet.

Noree · 05/24/08, 10:38 AM

Originally Posted by Ammanas

Well, with the new loot out I think its time to update the best-in-slot list.

Helm: [Dark Conjuror's Collar]
Neck [Amulet of Unfettered Magics]
Shoulders: [Mantle of the Malefic]
Chest: [Sunfire Robe] OR

11:11:41 called in wowhead_item::start:324 Item not found!

Back: [Tattered Cape of Antonidas]
Wrist: [Bracers of the Malefic]
Gloves:

11:11:42 called in wowhead_item::start:324 Item not found!

Belt: [Belt of the Malefic]
Legs: [Leggings of Calamity]
Feet: [Boots of the Malefic]
MH: [Sunflare]
OH: [Chronicle of Dark Secrets]
Wand: [Wand of the Demonsoul]
Ring: [Loop of Forged Power]
Ring: [Ring of Omnipotence]
Trinket: [The Skull of Gul'dan]
Trinket: [Shifting Naaru Sliver]

Gem [Runed Crimson Spinel] in red sockets and [Great Lionseye] in yellows until capped, then [Reckless Pyrestone]. [Glowing Shadowsong Amethyst] in the blue socket on helm/shoulders for the meta.

Like everyone else is saying, T6 shoulders are the piece to stay now which makes sense seeing as the new helm/gloves are off of KJ and ilvl 164 and the robe is a crafted 159 (not to mention T6 robe blows). Even with all of this new non-hit gear I'm finding Sunflare and hit gems to beat out the felmyst staff. Also, seeing as a lot of this is KJ loot there are a lot of other items that will still be very viable for most of SW (staff, sunfire gloves, even twins shoulders to an extent).

Will look better once the new items get added to Wowhead, but I'm sure you can figure out what the missing ones are. So am I missing anything here?

Yep this pretty much does it with the only exception being lack of crit. Using all Spell Damage, Hit and Haste gems you still stand at about 23% crit. Throwing in some Crit/SD gems in to reach near 25% instead would probably be better.

Wfatboy · 05/24/08, 12:00 PM

Any increase in haste rating is only realised if the caster is able to cast an extra spell in a given time period. If for any circumstance a cast is interrupted for the caster with higher haste, there will be no gain if that caster had not been able to acquire an additional cast relative to another caster with a lower haste rating.

The additional time required to give the caster with higher haste an additional cast is determined by (taking a time frame small enough to not need to life tap):
2.5X(Y+1) = 2.5Z(Y)

Where:
X = haste proportion, Z = haste proportion. (Z > X)
Y = number of casts
X = 1/(1+H), Z = 1/(1+H)
H = Haste % / 100 (Haste % given on character sheet)

The results of this equation need to be taken with consideration with respect to the situation you are dealing with.

More haste rating is realised (as damage) as your cast time decreases because the average amount of time wasted by a random interruption is less, therefore the addition of haste has driven your value of haste closer to its expected value. You can never have more effective haste than the expected value, as it assumes all of your haste rating has been converted to damage.
This is what I believe Daeven and Torq are alluding to, and their conclusions that warlocks with low levels of haste rating are not met with any realised gain in a majority of T6/T7 fights is correct. You can engineer your haste rating to give you extra casts much more often than not by using my equation.

Antoine · 05/24/08, 1:42 PM

Originally Posted by Noree

Yep this pretty much does it with the only exception being lack of crit. Using all Spell Damage, Hit and Haste gems you still stand at about 23% crit. Throwing in some Crit/SD gems in to reach near 25% instead would probably be better.

Putting it into Chardev, i got 21.16% tooltip crit which after talents is 29.16% destro crit, a very reasonable number which would probably put you around 30% raid-buffed. The rest of the stats, completely unbuffed (not even fel armor), in case anyone was curious, is 191 hit (capped with aura), 1368 shadow damage, and an incredible 467 haste rating which comes out to 29.63% haste. That would give you 1.81s shadowbolts.

Nas · 05/24/08, 4:55 PM

Originally Posted by Antoine

Putting it into Chardev, i got 21.16% tooltip crit which after talents is 29.16% destro crit, a very reasonable number which would probably put you around 30% raid-buffed. The rest of the stats, completely unbuffed (not even fel armor), in case anyone was curious, is 191 hit (capped with aura), 1368 shadow damage, and an incredible 467 haste rating which comes out to 29.63% haste. That would give you 1.81s shadowbolts.

Backlash is already incorprated into your tooltip crit, as it is to "all spells", Devastation isn't though. So 21.16% tooltip is 26.16% to Shadowbolt unbuffed.

Antoine · 05/24/08, 5:14 PM

My character on chardev had 0 talent points spent, so my previous statement remains correct.

It was a gnome though, so if you're any other race your crit will be lower probably.

PSGarak · 05/25/08, 1:12 AM

The spreadsheet value of haste is an average for a particular class of scenarios, not a best-case analysis. Consider hasting your shadowbolt from 2.5s to 2.4s and looking at a 24s time interval ended by whatever sort of effect you want. The unhasted warlock has time for 9.6 casts, but only casts 9. The hasted warlock fits in a 10th cast, netting an 11% improvement despite only 4% haste and a 4% spreadsheet increase. If you increase the window size to 25s, the unhasted warlock can cast 10, and the hasted warlock has time for 10.467 but only casts 10, gaining 0 DPS out of an expected 4%.

In summary: Yes, haste improves your DPS as a step function time rather than smoothly like your other DPS stats, with respect to time. However, this step function is NOT strictly below the smooth relaxation we model it as, and in fact the step function averages out to the smooth function. If you know exactly what time intervals you are dealing with you can optimize your haste rating for egregious effects. If you're looking at the average over several fights with several lengths and types of DPS interval, the spreadsheet is a good expectation value.