Idk

Yes, it's definitely a close call.. and probably not that important in the long run. However, it's important to me to squeeze out every last little bit of damage that I can. I realize that the circumstances of fights vary widely but that doesn't stop me from trying to make decisions as precise as I can.

I believe that spell haste is still a fairly poorly understood dynamic so I'm hoping that my investigation into its non-intuitive effects may shed some light on its usefulness outside of single-spell-spam.

 

Arelenda

I did read your post. It asked for thoughts.

I was merely pointing out that even for me, as destro, I barely value haste higher than +damage. For affliction it ought to be much, much lower due to only a few of the spells you use being affected by it.

To answer your question: I haven't done any modeling that's not already covered by the spreadsheet for affliction warlocks.

But here's a quick model:
Get 10% casting speed reduction. Shadow bolts are now 0.25s faster. On average, you'll gain half of that when refreshing a dot.

Gain on corruption: 0.125seconds average per 18seconds, or less than one percent. For longer dots it's worse. This is with 10% (about 175 haste) which is quite substantial.

Unless I'm misunderstanding what you're trying to say, this shows that dot uptime increase is negligible. 175+ damage would gain you around 7-10% extra damage, in comparison.

 

Morwen

With the current system the actual cast of UA should not be influenced by haste or latency at all (unless there is a spike). When you hit the UA key, a GCD will start from that moment, and it will end 1.5 seconds later at which point you can start casting again. The time at which the cast bar appears depends on latency; the length of the cast bar depends on haste; but due to the pseudo-queue mechanic neither of those should matter, you just hit the next cast as soon as your GCD is over.

My understanding of the cast mechanics is the following timeline:

0.0 - hit the key, GCD starts, cast request sent to server
0.0+latency - server acknowledges cast request
0.0+2*latency - cast bar pops up on client
1.5 - GCD completes on client, you are allowed to hit another key, so you do at time T>1.5
0.0+latency+castlength - cast completes on server
T+latency - server acknowledges second cast request. If this is "sufficiently close" to the event where the previous cast completes on the server, then this cast will get "queued" and will be considered to have been cast as soon as the actual moment the previous cast finished.

It feels like "sufficiently close" is currently +/- 100ms or so, the mechanics haven't been quite stable since the patch and I'm not aware of any recent rigorous tests on much exact room there is to hit the second cast on time. In any case if the timeline holds true then all that matters is T, which doesn't depend on latency or cast length, only on how frequently you physically hit the key.

 

galzohar

While the haste effect on dot uptime is small, I wouldn't call it ignoreable. If 10% haste increases DPS by 5% via increasing shadowbolt DPS and increases dot uptime by 1%, it's a total 6% dps increase instead of 5%. Also there's another second order correction of haste increasing the % shadowbolts are of your damage, thus the more haste you have the more effect it will have. Then again another correction is the more haste you have the more lifetaps/dark pacts you cast which means less benefit from haste... Overall you need to sum up everything to get something more accurate than the first-order "50% shadowbolt dmg therefore 10% haste is 5% DPS increase", which is probably not far off but from a spreadsheet prespective a 1% error is not neglicible.

 

Trickykid

If we take the case where every time all DoTs and Curses are up a Shadow Bolt is started, then close to every time a DoT or Curse expires, you will be in the middle of a Shadow Bolt cast. Given a long enough fight the distribution of those expirations should fall evenly across all parts of the Shadow Bolt, so each DoT on average suffers half a shadow bolt (plus lag) of downtime per expiration.

To make the spreadsheet take this into account for "the next stat" section, simply change the 'avg dot gap' to =D23/2 on both the 'DPS' and 'TNS-haste' sheets.

Note: This is the maximum gap that occurs due to shadow bolt overlap. That gap is shortened if the player chooses to LT/DP during some windows to minimize the downtime.

Responding to some of the previous posts: just because haste may be less useful to an affliction warlock doesn't mean they shouldn't understand the value it gives.

 

Idk

Like I said earlier, I'm not disputing that spell damage is better than spell haste. I'm trying to find out how good 1 spell haste is relative to 1 spell crit or 1 spell damage or 1 spell hit or 1 mp5 or 1 int. All of these stats matter and every piece of gear has a mix of at least some of these damage influencing stats. I can't choose gear with 100% spell damage and nothing else, so I have to understand how effective the other stats are at increasing my damage.

Let me say one more time for emphasis.. I'm NOT saying that haste is better than damage or hit. I'm saying the spreadsheet is not accounting for all of the ways that spell haste CAN improve damage and therefore undervalues its contribution to overall damage.

Exactly the reason why I'm modeling this and trying to figure it out.. I believe the spreadsheet is wrong and I believe my model backs this claim.

What you fail to realize is that "less than 1%" damage boost for the dots is in addition to the contribution of additional shadowbolts due to spell haste. The spreadsheet is only counting additional shadow bolts.. therefore it's undervaluing haste. To accurately understand how valuable spell haste is, we need to be able to include both effects.

Here's a similar situation.. assume the spreadsheet didn't actually change the ISB uptime when you added additional crit. The spreadsheet would tell you that crit improves DPS by less than it actually does because it's not factoring in that additional ISB time. It doesn't mean that crit is better than spell damage or spell hit or whatever.. it just means the spreadsheet is incorrectly accounting for the mechanic.

 

Idk

First off, thank you! Your last sentence is perfect.

In general I think your spreadsheet changes are an appropriate approximation of the effect.

I do believe, however, that your assumption "Given a long enough fight the distribution of those expirations should fall evenly across all parts of the Shadow Bolt" is incorrect. The reason is that even though there are random effects like resists, the limited number of actions that can be performed all have regular cast times and intervals between cast times (assuming lag isn't in the picture). The GCD is always 1.5s, the UA cast time is always 1.5s, the UA and Corr durations are always 18s, the SL duration is always 30s, and the SB cast is a fixed time based on your spell haste. Since the chunks are consistent (and all happen to be multiples of .5 except the shadowbolt) I do believe that certain cast times of shadowbolt will fit into the gaps better than others. I don't have the math to support this but I do believe that my model demonstrates it. Note how dot uptime for the 18s spells increases from 0% haste to 5% and drops down by 6%.. then increases again up to 10%.

Is this way too theoretical to actually be useful in determining real world damage? Probably... but I do find it interesting.

 

Trickykid

The gaps won't be consistent given that the lag on top of every cast varies constantly. As soon as haste is involved, the gaps become less predictable as well. You can go ahead and try to model it more with your simulation, but simulations can create bizarre results if you simplify things too much. You are seeing a downgrade by going from 5% haste to 6% haste, which highlights that there are some problems. If you want to make your claims more robust, I would include more 'real world' elements like LT/DPs and a randomized lag value.

 

PSGarak

Everything is a multiple of 0.5, but with a variable DoT order which multiple of 0.5 you're going to be displaced by varies considerably--IF the 2.3 casting mechanics mean that everything lines up nice like that in the first place. Being uniformly distributed across the 5 multiples of 0.5 is statistically equivalent to being uniformly distributed across the entire interval for our purposes. And of course, as trickykid said, as soon as you add haste in there everything goes to pot and you basically end up with a continuous uniform distribution regardless of lag mechanics and casting cycle.

However, that's assuming a naive casting cycle, which is what I think Idk was actually getting at. If your mana pool is not within one lifetap of either full or empty, you can spread around your lifetaps strategically to decrease DoT-gap. Specifically, if one of your DoTs is falling off you can choose to either DoT now and have an action up in 1.5s followed by a shadowbolt, or shadowbolt now, have an action in 2.5s, and lifetap afterwards. Both orders end up with the same spells cast and net man cost, but one of them nudges the DoT up by 1.0 seconds. You can also nudge the DoT back by 1.0 seconds if it wouldn't be quite done following a lifetap.
So while a constant lifetap/shadowbolt order will pretty much always have a DoT-gap of one-half your average spell cast time (average of spell cast times weighted by spell cast measure), that's more of a worst-case scenario which a skilled warlock can improved on by up to 0.5 seconds on average by forecasting 2.5 seconds and planning shadowbolts/lifetaps accordingly. It might be possible to further improve that by forecasting further ahead than one tick or one cast (ie two lifetaps vs 1 shadowbolt to nudge the action timing by 0.5 seconds) but the returns on doing so drop off after about three GCDs of forecasting (by which point you can control to the 0.5s), and larger forecast horizons requires a larger and larger mana buffer on both ends, increasing linearly.

Assuming you're at 50% health, 50% mana, and completely flexible in your cast cycle (ie lifetap/DoT distribution not dictated by movement) you should be able to control within 0.5s when you refresh your DoTs for an averge gap of 0.25s while maintaining steady-state dps, unless two DoTs overlap which can't be strategized out of. Once you add haste into the equation the dynamics of forecasting get a lot messier because things don't line up as nice. In theory it might lower your DoT-gap under ideal circumstances but the calculations become disgusting.

http://sig.gamerdna.com/quizzes/INFL...erks/Garak.png

 

Arelenda

Well, yes, of course.

You claimed the spreadsheet doesn't model haste well, because it only models added SBs, it does not compute increased uptime.

I calculated that this uptime gain is significantly less than 1% for even an absurd amount of haste like 175. I considered this fairly insignificant compared to the current calculated gain from haste. You disagree with this last bit.


Can we agree on these points?

- yes, it's missing.
- It's very complex to model or calculate, for small values player timing/lag is obviously much more significant
- when modeled with a high amount of haste we get a very small gain in dot uptime
- affliction warlocks will not try to stack haste, but will occasionally get it anyway because loot drops are random.

In order to answer your question: no, the spreadsheet doesn't model it.

Haste does a lot of things a spreadsheet can't model anyway: faster fears, banishes, and spells in general. For example, even though immolate/UA is GCD capped, having haste makes you finish them faster, allowing you to move sooner. We can add "increases dot uptime" to the list.

Players should value it a bit higher than what the spreadsheet says, because of this.

 

Idk

Yes, we can agree on those points. What I don't agree with is the desire to simply ignore this effect on dot uptime because you think it's insignificant or hard to model. You can take the latter 3 statements from above and replace "dot uptime" with "ISB uptime" and "haste" with "crit" and the statements are just as true. ISB is very complex to model or calculate, it's significantly affected by timing and lag, crit increases will produce a very small gain in ISB uptime, and affliction locks will not try to stack crit but will occasionally get it anyway. Does this mean it's not important to model the effect of crit on ISB uptime? Of course not! Considerable energy has been put in to understanding ISB uptime even though it's extremely hard to model across a raid of players. Our models are approximations at best but at least there's been an attempt to create mathematically sound models of ISB uptime rather than making unsubstantiated assumptions.

I'm not saying that this problem requires an equal amount of effort as the ISB uptime models, but I don't understand why it's being dismissed as insignificant without further investigation.

My model shows me that haste has a pronounced effect on dot uptime that goes much further than reducing the average dot gap by 50% of the cast time reduction of shadowbolt. I want to be SHOWN why my model is wrong.. not simply told that it is without any evidence to support it.

I have a set of gear with 4.88% haste that i use when I'm playing destro. I'll try to do some Dr. Boom testing to see if my model comes even close to real world testing.

 

Arelenda

Unless I'm missing the point again, you're arguing that haste will do the following:

- allows you to get Shadow Bolts in. We agree on this. It is already modeled.
- allows you to casts finished quicker, which can be very beneficial on mobile fights or other tasks. We agree on this. It is not modeled.
- will allow you to alter your spell choice so you can cast more dots instead of Shadow Bolts, increasing your damage per casting time. We agree on this, but are disputing the significance of this particular factor.

Would it be hard to add any of the following to the damage simulation?

- lifetaps/darkpact
- Immolate (unless you're not using this because of lower dps, but make sure to state so)
- damage output
- haste rating instead of hasted casting speed (as this is not linear, statwise)

This would at least give you a rough estimate of how far the spreadsheet is off.

 

Bolche

I first thought of making 'avg dot gap' to something like 'filler cast time'/2, but it does not factor when two dots expire at the same time and you can only renew one at a time and dot refresh resist. Moreover, I wanted to let users adjust manualy something related to the player attention. I know for exemple that when I was affliction, my real average dot gap (from WWS) was something around 3-4s under ideal conditions, like Al'ar p1. During fights whith many interruptions, it could go very high. That's why I let it up to the user.

But it is obvious that the cast time or your filler spell (and thus haste) as an impact on your dot uptime. When tanking and spaming searing pain, I know that (once aggro is OK) I can maintain an almost perferct dot uptime, thanks to the 1.5s cast time of SP.

It seems logical that reducing the cast time by X second does reduce the average dot gap by X/2. So to take into account in this in "the next stat" section while leting the user choose what it wants for the dot gap, I can set the dot gap in the TNS-haste section to : dot_gap - 'filler cast time reduction'/2
In the .xls doc it translates into : =DPS!J10-(DPS!E24-E24)/2


By the way, if you have some suggestions for the spreadsheet, please post them into the dedicated topic (http://elitistjerks.com/f31/t10065-a...s_spreadsheet/).

 

Idk

This is doable, I'll see what I can do today.

As far as excluding immolate, that was a personal decision based on the fact that I rarely use it.. in the raids that I regularly run I'm the only warlock, there's a shadow priest, and no mages. I just can't justify the cast. I can, however, include immolate in the model quite easily. My plan is to make it the lowest priority dot.

 

Arelenda

Thanks, let us know the results.

Even without CoE/Scorch and with CoS and Shadow Weaving, it might do more damage-per-casting-time than SB, with your spec. Just plug it in the spreadsheet or run ShadowSeer and see what comes out. If it's close enough to Shadow Bolt, my guess is you can use it to improve on DPS since it's a 1.5s cast option, usable to improve dot uptime for reasons you stated earlier.

 

Idk

I've extended the functionality of my model to include:

- immolate
- lifetap (I arbitrarily chose 2000 mana per tap and 8000 max mana)
- reduced casting time of immo and ua with haste (even though GCD remains the same)
- changed from haste to haste rating using the following formula (let me know if this is incorrect):

base cast / (1.0 + (haste_rating/1570)) + lag

The really fascinating thing that has come out of this is seeing how lifetap affects the casting timeline. I've tried a variety of methods to improve general dot uptime, with some being more successful than others. This is what I've settled on for the moment:

If we have less than 1000 mana, lifetap (I sampled with 500, 1000, and 1500 for the "dangerously low on mana" threshold.. and found 1000 to be the best).
If UA isn't up, cast it
If UA is going to be up in less than ua cast + lag time, recast it
If Corr isn't up, cast it
If SL isn't up, cast it
If Immo isn't up, cast it
If Immo is going to be up in less than immo cast + lag time, recast it
Consider a preemptive lifetap: if any dot has more duration left than that dot's cast time (or the gcd) but less duration than the cast of a shadowbolt AND the current mana pool is more than 1 lifetap from being full, cast a preemptive lifetap.
Otherwise, cast a shadowbolt

I'm now cycling through haste rating from 0 to 105 in increments of 15 (arbitrary numbers) trying to generate enough data to feel confident. What I find interesting is that the variability between simulations of the same haste rating is very small. I think this is a good thing.

 

Arelenda

Looks promising. Keep us posted. Maybe try some variations with lower hit chance as well, and see how it affects numbers?

 

manupod

Ok I have a couple questions to ask of you guys & gals. Just to touch on some of the Curse & Trinket discussions.

Ok the first thing I'd love to have cleared up is about trinkets in relation to when their effects are applied or available to be applied to spells.

Example: I open up a fight like Lurker with Amplify Curse > CoD > Pop Icon of the Silver Cresent > UA > Corruption > Immolate > SB SB SB and so forth. (Assuming I'm not on CoE/S duty of course)

What's confusing is the trinket use. I understand (maybe wrongly) that trinkets don't effect AC and CoD. So am I right by doing my CoD first?

My second thing I'm confused about is CoE/S use. I usually apply CoE/S before applying DoTs. I do it for two reasons, but here's the reason I'm not sure makes sense: CoE/S won't affect my own DoTs if my DoTs went up before the Curse. Is that true? Or does it not matter? I know it doesn't matter too much unless I'm being min/max about it but it's just been bugging me and I figured I'd ask.

Any input is appreciated. And apologies in advance if this is breaking any flow of conversation going on.

 

galzohar

Why would trinket not affect CoD? Afaik CoD is affected by spell damage, and trinket is no different. Spell damage effect is determined when the spell leaves your hands. Why do you CoD though? As affliction you *should* be the malediction "bitch" with 13% CoS let a destruction warlock do CoD since he doesn't have malediction (or better, CoR and CoE so only CoD if you have more than 3 warlocks in a raid and/or don't have at least 2 fire/frost mages and/or the fight does not allow use of CoR).

Afaik CoS needs to be active when your DoT ticks, so casting it right after the DoTs should work. Remember though if at least 2 other shadow users started to DPS at least 1.5s before you did, CoS first will yield more raid DPS. In optimal situations though the curse should come after the UA+corruption.

 

PSGarak

CoD is affected by trinkets. Since it's coefficient is so large (200%) it actually gets the most use out of your trinket than any of your spells. While this coefficient is small relative to its duration that has no effect on how much extra damage you get out of the crescent use.

DoT spell damage from gear and caster buffs gets locked in on cast, but mob debuffs are reevaluated each tick. Assuming every other shadow caster in your raid also leads off with DoTs you can increase your DoT uptime, and therefore damage, by doing UA corr CoS, with no disadvantage. Longer than that and you start having DoTs tick without it.

http://sig.gamerdna.com/quizzes/INFL...erks/Garak.png

 

Arelenda

CoD gets 200% bonus, more than any other spell. Trinkets are very beneficial with CoD.
CoD damage = 4200 + 2 * spellpower.

I think you're confused because amp curse only affects the base damage. So trinkets don't affect the boost given by Amplify Curse:
Amped CoD = 6300 + 2 * spellpower.





Note that CoD is not affected by Shadow Mastery, for reasons unknown.

 

manupod

Great replies there, and thank you for clearing that up. Of course I CoS/E most of the time, but the example I gave with CoD was just that, an example. Don't think I'm greedy and CoD all the time =)

I think it's best explained by exactly what you said, where it's re-evaluated every DoT tick. I'll definitely be changing up some of my rotations for certain fights now.

Again, thanks a bunch!

Now back to your regularly scheduled discussion...

 

t0qt4u

Is the Warlock DPS Calculator reliable? If so, I've tested most of the specs through it, and am curious to know if someone can find a spec with the most DPS.

 

Idk

EDIT: added some damage numbers to get an idea of overall change of damage.

Results from my model. They're interesting.. still not quite sure what to make of them. To recap:

Assumptions:

8000 mana
lifetap returns 2000 mana
no cataclysm
.1s of lag
no curses or soulshatter
1% resist chance
600s of modeled combat
stand and cast (no movement)
~850 simulated combats per chunk of haste rating (this model is much slower to calculate than the old one)

Picked some approximate damage numbers, these could be more precise I'm sure.. but since they stay constant across all measures of haste they should be a decent approximation of damage difference from haste:
ua: 4000 damage
co: 4000 damage
sl: 3000 damage
immo: 1500 damage
sb: 3000 damage

Priority of actions:

If we have less than 1000 mana, lifetap (I sampled with 500, 1000, and 1500 for the "dangerously low on mana" threshold.. and found 1000 to be the best).
If UA isn't up, cast it
If UA is going to be up in less than ua cast + lag time, recast it
If Corr isn't up, cast it
If SL isn't up, cast it
If Immo isn't up, cast it
If Immo is going to be up in less than immo cast + lag time, recast it
Consider a preemptive lifetap: if any dot has more duration left than that dot's cast time (or the gcd) but less duration than the cast of a shadowbolt AND the current mana pool is more than 1 lifetap from being full, cast a preemptive lifetap.
Otherwise, cast a shadowbolt

The format of the results:
Haste: haste rating
UA up: percentage of unstable affliction uptime
Co up: percentage of corruption uptime
SL up: percentage of siphon life uptime
Im up: percentage of immolate uptime
SBs: number of shadowbolts cast
LTs: number of life taps
preemp: percentage of life taps that were cast preemptively.. that is, lifetaps that were used to reduce the time between dot expiration and dot refresh where a shadowbolt would have caused a larger gap
needed: percentage of life taps that were cast because mana was less than 1000
total: total damage done
change: ratio of total damage done with this amount of haste vs. total damage with no haste

The results for haste rating of 0 to 157 in increments of 15.7 (so it should be approx. 0-10%):

---

Haste: 0.0
UA up: 94.8
Co up: 94.5
SL up: 95.5
Im up: 86.0
SBs: 127.0
LTs: 48.0
preemp: 57.3
needed: 42.7
total: 742300.0
change: 1.0

Haste: 15.7
UA up: 95.5
Co up: 95.2
SL up: 94.6
Im up: 86.9
SBs: 127.0
LTs: 48.0
preemp: 71.3
needed: 28.7
total: 744167.0
change: 1.0025

Haste: 31.4
UA up: 96.0
Co up: 95.7
SL up: 95.0
Im up: 87.4
SBs: 128.0
LTs: 49.0
preemp: 71.4
needed: 28.6
total: 749040.0
change: 1.0091

Haste: 47.1
UA up: 95.5
Co up: 95.4
SL up: 94.8
Im up: 83.7
SBs: 131.0
LTs: 49.0
preemp: 51.9
needed: 48.1
total: 754633.0
change: 1.0166

Haste: 62.8
UA up: 95.9
Co up: 95.8
SL up: 95.2
Im up: 84.2
SBs: 131.0
LTs: 49.0
preemp: 51.3
needed: 48.7
total: 756240.0
change: 1.0188

Haste: 78.5
UA up: 93.0
Co up: 92.7
SL up: 96.4
Im up: 85.9
SBs: 133.0
LTs: 49.0
preemp: 67.3
needed: 32.7
total: 755980.0
change: 1.0184

Haste: 94.2
UA up: 93.6
Co up: 93.4
SL up: 96.4
Im up: 80.3
SBs: 134.0
LTs: 49.0
preemp: 53.0
needed: 47.0
total: 757353.0
change: 1.0203

Haste: 109.9
UA up: 93.6
Co up: 93.5
SL up: 94.6
Im up: 87.1
SBs: 135.0
LTs: 50.0
preemp: 51.0
needed: 49.0
total: 763487.0
change: 1.0285

Haste: 125.6
UA up: 94.1
Co up: 93.9
SL up: 95.0
Im up: 87.2
SBs: 135.0
LTs: 50.0
preemp: 50.5
needed: 49.5
total: 764987.0
change: 1.0306

Haste: 141.3
UA up: 94.5
Co up: 94.4
SL up: 95.4
Im up: 87.7
SBs: 136.0
LTs: 50.0
preemp: 50.7
needed: 49.3
total: 769727.0
change: 1.0369

Haste: 157.0
UA up: 95.0
Co up: 94.9
SL up: 95.9
Im up: 88.4
SBs: 136.0
LTs: 50.0
preemp: 50.7
needed: 49.3
total: 771780.0
change: 1.0397

 

Arelenda

These do seem a bit dodgy. They will actually affect your results, because there is usually no reason for you to cast Immolate if has less damage per casting time than Shadow Bolt. Also:

Immolate does 942 base and gets 85% from spellpower.
Shadow Bolt does 575 base and gets 85.7% from spellpower.

There is no way one can do 1500 and the other 3000. Even when you account for crits, and the fact that SB gets 10% extra damage through Shadow Mastery. Not even with no Coe/Imp Scorch and Shadoweaving.

I'd recommend going in the game and tossing each of these spells on a target and see how much they do and using those values. I'll also PM you some bits of code that have coefficients, if you're so inclined. It would allow you to determine the dps increase given by extra spellpower, too, in your simulations.

8000 mana seems very low, but that should hardly matter. 10000 would probably be more realistic in a raid scenario, where you have AI and MOTW.


So 157 haste gets you almost 4% more damage. That's quite good. +157 damage would have netted you around 8%, I'd wager. Roughly 2 haste rating = 1 spell power. Interesting.