With all the talk about cooldown stacking TC lately I feel I should explain more in detail how Rawr.Mage works as one of its main purposes is to answer this very question. Many of you will remember the http://elitistjerks.com/f40/t14250-m...ell_selection/ posted some time ago which showed that assuming we have only mana constraint then it is never optimal to use more than 2 cycles. What Rawr.Mage uses is a generalization of that framework extended to apply to a more general problem.
At the core is the problem formulation. In a given fight at any point in time we can choose to follow any number of algorithms for spell selection. A cycle is one such fully specified algorith. For example a cycle can be casting Fireball until Scorch debuff is almost out and then Scorch. Another cycle can be AB-AM-AB-AM-AB-AM-FrB with AM following AM if AM procs clearcast and as many Frostbolts as needed to not drop AB debuff. Or it can just be Frostbolt spam. One important thing is that they have to be fully specified, meaning at any point in time while executing the cycle it should tell what to cast next. Given such specification it is then possible to compute long-term average dps and mps of the cycle (and possibly other relevant values such as tps).
We also have a number of cooldowns. The way I'm treating them is in the same way that Manly specified. It is an ability that changes the dps/mps values of cycles and is controlled in a sense that we can either choose when to use it or we know when/how it is used. At any point in time now we have to decide which cooldowns should be active (or we know are active) and which cycle to use. While we can use any cycle at any point in time, this is not true for cooldowns. There are constraints imposed on them. This is the key change compared to the 2-cycle theorem that can cause more than 2 cycles to be optimal.
We can now formally specify our problem. Lets enumerate all combinations of cooldown combinations and cycles and let be the number of all cooldowns+cycle combinations. For example number 143 could corresponds to Combustion+Flame Cap+Destruction Potion+Molten Fury+Heroism+FireballScorch. Each combination of cooldowns and cycle is assigned its own number. Let now be ammount of time spent casting cycle associated with number i while cooldowns associated with i are active. We also have special variables reserved to represent the use of mana potions, mana gems and evocation. While they have a special meaning they can be treated the same way as other cycles in terms of the problem formulation.
What we want to optimize then is total damage done:
This is subject to a number of constraints. I'll show here some examples so that you get a general idea. First we have duration constraint:
Then we have mana constraint:
Then we have a constraint for potions (for the purposes of formulation we assume mana potions take a certain small ammount of time to use):
A trinket constraint might be:
where maximum activation is computed as .
Similarly we have constraints that tell maximum time spent evocating, time spent in heroism, time spent in molten fury. In addition we need constraints on cooldown stacking. For example even if we can use trinket twice we can not use it both times during molten fury. In the current implementation there are 39 constraints and more might be added to fine tune some edge cooldown stacking requirements.
The important thing is that this gives a theoretical framework under which we can search for optimum solution. What Rawr.Mage does is it looks at all possible combinations of cooldowns and cycles and solves this problem and returns the solution that optimizes total damage done given the constraints. There is one restriction, which is that constraints have to be linear. Some cooldown stacking constraints are harder to express in linear terms so in some cases some decisions have to be made. Currently the constraints are only added if there is no case where the constraint could be violated. This means that in some cases the solutions can be slightly overestimated. An alternative would be to add overrestricting constraints that would guarantee that all stacking solutions are possible, but might also removes some, this giving us a lower bound. That would give us an error margin to the calculation so that you know in which range the true optimum solution is.
The conclusion should be that solving cooldown stacking is a hard problem to do ad hoc. You can give some heuristics as to what cooldowns are better to stack and which give more value. But in the end if you want to take into account all interconnections you have to specify the problem in general terms as was shown here. Then it will automatically take into consideration all cooldown interactions. It will know how much Destruction Potion hurts Combustion, how much clipping of AB under Heroism+IV costs. It is working with values of average dps and mps of all cooldown and cycle combinations and balancing them against all constraints. When you get a solution you know that it is optimal under those constraints. If it is using AB with Heroism+IV and it is going below 1 second it is doing that because in the big bicture that gives more total damage than stacking cooldowns some other way.
So enter your gear and let Rawr.Mage do the theorycrafting for you.
Also are there any plans to add unpredictable fight duration? A 4m fight and a 3.5~4.5 minute fight could have different results especially when mana is taken into consideration. Maybe it would just be better to maximize your DPS for the "worst-case-scenario" fight duration at least on fights that are not on farm, since if it ended early you still win, but then there are lots fights where it's very hard to put your finger on their exact duration.
Well, debatably that could be dumbed down to 'use a playstyle that errs towards DPM for the first 80% * of the fight, then use a different playstyle for the latter part to adapt to a varying fight duration'.
* 80% would vary depending on the fight duration elasticity.
<Eej> YOU"RE GONNA PULL
<Eej> IF YOU SQUEEZE OFF ANOTHER ARCANE BLAST
<Spectear> You've obviously never played with Manly.
<Spectear> That's hardly a reason to stop DPS. Very Manly Staff
The constraints that are hard to express have to do with packing and scheduling. For example if you have typical trinket with 20 sec activation and 2 min cooldown then on a 5 min fight you can use it at most 3 times or total of 60 seconds. A max 20 sec can be used during heroism, max 40 sec can be used during IV, max 30 sec during AP.
It meets all those constraints but it is clear you cannot split them into 3 activations of 20 sec. You could probably come up with some constraint that would prevent this from happening but it feels like there would have to be a lot of them, discarding these kind of scenarios almost on a case by case basis. If anyone comes up with a general solution I'd be very interested to hear it. It just feels that to account for all cases you'd have to treat the problem as a general scheduling optimization which is computationally more prohibitive.
Regarding variable fight duration I don't have any plans to add it at the moment. As Manly suggested you can just examine the cooldown placement and spell selection for a range of fight durations. In most situations the transitions are gradual and the only change that happens is percentage of time spent in different cycles.
My apologies if this was covered already and I failed to find it, but has there been any considerable theorizing on how the new Sorcerers Alchemist Stone can affect the Mage's overall damage in longer fights? With the addition of many more pieces of desirable high haste gear, might the stone become one of the more sought after trinkets to prevent any kind of downtime in damage per second due to having to evocate or otherwise regain mana in some way, especially when choosing Molten Armor over Mage Armor?
A while back someone, Manly I believe, posted that he often considered stopping dps near a bosses death to let his ignites tick off. This is ostensibly to remove the possibility of resetting the ignite timer with another crit (especially since the ignite could be a rolling one whose damage is greater than that of a crit fireball). I've been thinking that a better alternative would be a frostbolt -> ice lance/frostbolt or an arcane blast (-> arcane blast). I suggest this especially considering that with haste, most likely passive at the point of boss death, multiple casts would be possible in this window. The difficulty of course being able to feel whether or not the boss will die before ignite ticks off (where another fireball would have been more prudent). And of course this hinges on the ignite dot being added to the mob, or refreshed, at a very precise time
For example:
Boss dies at 4:00 so at 3:55 the last fireball is cast.
Immediately the player begins casting a Frostbolt or Arcane Blast
Depending on haste, another cast between 3:57 and 3:58
-or-
An instant such as fireblast or ice lance is cast
This being my first post here, I'd like to get some thoughts on what people think of this, such as:
Is this viable to try, or a waste of effort to secure a small or possibly non-existent damage increase?
A lack of haste (or the presence of it, for that matter) would shift the scales more towards a double arcane blast. Do buffs such as Imp CoE and Winter's Chill shift the balance decidedly towards frostbolt?
Does frostbolt's travel time automatically void the argument in favor of Arcane Blast?
What kind of damage increase -- and how relevant to actually killing a boss -- would this be?
Yes, as you pointed out, frostbolt is immediately out because of travel time. The problem is that there isn't a lot of alternatives. The contenders are:
-Arcane blast (with the major issue that the first cast is far too long for what we're looking for)
-scorch
-fireblast
-dragon's breath
-blastwave
I realise that if dragon's breath/blastwave is in that list then it implies melee range, in which case frostbolt would be equally a contender. But to be honest I don't think any of them is realistic. This pretty much only leaves us with scorch and fireblast. They both have the annoying 'feature' of possibly delaying further the ignite tick, which is exactly what we're trying to avoid. I will point out a technique I use on trash, but that I have never gotten around to do on bosses. As everyone knows, you can get more than 3 crits with combustion, possibly even a near infinite amount of them. I have always been against doing fireball/fireblast for the 3rd/4th combustion crit, since I know it causes ignites to be overriden. However, this only applies for as long as you plan your ignites to tick. Also, this 'rule' only applies when doing single targets. There is nothing wrong with using your 4th combustion crit on a secondary target, since there is no loss involved. It also means that you might as well go for that 4th combustion crit if you know your target will die before the 2 ignites ticks get to tick. Although, and this is the tricky part, in practice I would have a hard time conceiving a case where I would pop combustion that late, precisely when I have a huge amassed ignite, to be in that kind of boss hp%. It would be the kind of case I would want to avoid getting in in the first place. But, the same principle applies here, to a level, even if combustion is not up. What if were better off finishing a boss with scorch/fireblasts ? I already to that on every trash mob, a boss should be the same. I think in the ends it becomes a judgement call, whether or not you have a huge rolling ignite that you don't want to risk letting fall off. This is really calling a long shot.
Personally most of the time I finish with fireblast/scorch spam for the last 5 seconds or so. I fully realise that this is working against the whole 'dont let ignite fall off' debate, but I am not sure there is any clean solution to that problem. I think arcane blast is not suited - at all - for the task. The major issue is that it has piss-poor DPS on the first 2 casts, and our goal is solely to provide dps for 4 seconds of time. Sure, its our best non-ignite-proccing alternative, but I would doubt its dps is anywhere near anything acceptable.
<Eej> YOU"RE GONNA PULL
<Eej> IF YOU SQUEEZE OFF ANOTHER ARCANE BLAST
<Spectear> You've obviously never played with Manly.
<Spectear> That's hardly a reason to stop DPS. Very Manly Staff
If you've been pushing the ignites back couldn't you, at say the 6 second mark (till boss death) simply wait an extra 300 ms or so to cast your fireball putting it over 2s by a decent margin?
Assuming that you have a respectable amount of passive haste you'd have to wait longer than .3 seconds in order to assure yourself that a crit fireball won't be resetting ignite.
And to Manly, the random nature of the crit just before a boss death makes it unavoidable. I'm attempting to gauge whether an 8k crit fireball and a fireblast would be more damage than, say, 2 1.6k ignite ticks, 2 ABs and a Fireblast, or 2-3 scorches and a fireblast.
I disagree with ruling out AB due to the painful initial cast time. Again considering this is a person judging when a boss is going to drop from 1% to 0% (or watching actual numbers drop using a mobhealth addon) they are open to misjudging how much time is left in the encounter as well as the dps other members of the raid are putting out in this area. The pros of AB are the ability to do something other than stand still without risking wasted ignite damage with another fire crit. When the possibility of poor "time to die" judgment is factored in AB is more forgiving due to faster subsequent casts (which rewards spamming in the event that the boss takes longer to die than estimated). As you say the risks involved in scorch -> fireblast are what we're trying to avoid.
What the question boils down to is "What is the best use of 5 seconds of dps time, assuming you go in with a full ignite dot ?" Note that we have to assume that this isn't a snapshot of a fight, but the end of one, and that the time isn't limited to 5 seconds, but can go over or end abruptly.
1. 0:00- Flame cap/Destro pot/IV/Combustion/Skull
2. 0:20- Icon/head
3. 2:00- Mana Pot/Skull
4. 2:20- Icon
5. 3:00- Gem
6. 4:00- Nothing! i'm retarded put a number here
7. 5:00- Flame Cap
8. 5:20- Icon(Not skull yet)
9. 5:40- Skull, Destro/IV/Combustion
Hero Should be at 5:15ish.
Kind of backtracking this thread by 2 or 3 pages, but say I don't have a Skull (which I don't, HSH and Icon are my 2), and assuming you would not want to pop everything at 0:00 for the sake of scorches (Unless you can sucker another mage into doing 5 x scorch and then keep it up) I'm going to say it should look more like this (also assumes at least 1 other Fire mage):
Kind of backtracking this thread by 2 or 3 pages, but say I don't have a Skull (which I don't, HSH and Icon are my 2), and assuming you would not want to pop everything at 0:00 for the sake of scorches (Unless you can sucker another mage into doing 5 x scorch and then keep it up) I'm going to say it should look more like this (also assumes at least 1 other Fire mage):
* Popped as soon as MF range/Heroism is up
+ 20 seconds after Icon is popped, to allow for HSH
0:03 Seems a tad aggressive to pop everything.
You're only getting 3 trinkets regardless and two icey veins so you have some room to pop them strategicly.
Pop the cap on pull, but save the destro/iv/comb/head for 0:30 or so right after you refresh scorch.
That's true. In doing the above time table, I'd snag a couple 6k+ crits from the get go and subsequently never live down ripping threat within the first 10 seconds of the fight. Given the time frame, I can be a little more flexible with my cooldown popping.
Thanks.
What I've usually done is pop cooldowns on the first tank switch, about 30 seconds in. It gives a good reference frame to coordinate with other group members (for the first drums if you have only one or two LW in group, for example).
If you've been pushing the ignites back couldn't you, at say the 6 second mark (till boss death) simply wait an extra 300 ms or so to cast your fireball putting it over 2s by a decent margin?
Originally Posted by Actovision
Assuming that you have a respectable amount of passive haste you'd have to wait longer than .3 seconds in order to assure yourself that a crit fireball won't be resetting ignite.
And to Manly, the random nature of the crit just before a boss death makes it unavoidable. I'm attempting to gauge whether an 8k crit fireball and a fireblast would be more damage than, say, 2 1.6k ignite ticks, 2 ABs and a Fireblast, or 2-3 scorches and a fireblast.
I disagree with ruling out AB due to the painful initial cast time. Again considering this is a person judging when a boss is going to drop from 1% to 0% (or watching actual numbers drop using a mobhealth addon) they are open to misjudging how much time is left in the encounter as well as the dps other members of the raid are putting out in this area. The pros of AB are the ability to do something other than stand still without risking wasted ignite damage with another fire crit. When the possibility of poor "time to die" judgment is factored in AB is more forgiving due to faster subsequent casts (which rewards spamming in the event that the boss takes longer to die than estimated). As you say the risks involved in scorch -> fireblast are what we're trying to avoid.
What the question boils down to is "What is the best use of 5 seconds of dps time, assuming you go in with a full ignite dot ?" Note that we have to assume that this isn't a snapshot of a fight, but the end of one, and that the time isn't limited to 5 seconds, but can go over or end abruptly.
Edited for clarification.
I agree with the ideology of considering 300ms of no dps time in order to avoid having your ignites fall off, however, I feel compelled to point out the nature of the problem. You do not know when the boss is going to die. It could be < 300ms. You can get a mini-guesstimate, but its very hard to do becuase you can't realistically tell the difference between T-minus-5 and T-minus-3, which is where the entire problem resides. This is also why I ruled out AB. I think any decent raider knows when the boss is between T-minus-10 and T-minus-5, but once you get below that point its very hard to guesstimate. To be more to the point; I believe that only 1.5s or less casts are to be considered for cases where 5 or 3 seconds left on a boss. You would not want to cast a 2.5s frostbolt, since your odds of having that 2nd frostbolt not land are too high, and you additionally have the odds of your travel time to screw you on the first frostbolt. And since my example case *specifically* tries to deal with T-5 or less cases, I don't think AB is a good contender. The ramp-up time, in addition to having AB do 2.5, 2.5, 2.2, etc. is really going to kill it. The only alternative, if you want to consider AB, is to consider, literally, AB spamming since the *start* of 1%, and not really gun for the very very end of the boss HP. This in itself would require math to consider. I doubt the results will be favorable in any case.
And I must regretfully point out that time-to-die is woefully inadequate for handling specifically that case. The way the mod works is to record the times where the boss HP, in %, changed. Since it is the only information available. The mod then 'graphs'/extrapolates an estimate of how fast the boss will die, but all of this assumes a constant DPS. It totally fails a taking into account simple things like bloodlust, or execute range. The mod is somewhat accurate for 100-20% boss HP, but past that point its not much reliable. And even so, you can't either rely on the timings from the start of time to die (ie: 50% boss hp+) since those data points also ignore the same things (bloodlust/execute range). This means at 50% it might tell you that you have 3:30 left, but in practice you have 2:58 left. Too bad you used icy veins so soon.
<Eej> YOU"RE GONNA PULL
<Eej> IF YOU SQUEEZE OFF ANOTHER ARCANE BLAST
<Spectear> You've obviously never played with Manly.
<Spectear> That's hardly a reason to stop DPS. Very Manly Staff
All this would be so much easier if we got a Conflagrate-like skill that blows any amount of ignite left and adds a tad on top... Are you listening, Blizzard? ARE YOU?
And I must regretfully point out that time-to-die is woefully inadequate for handling specifically that case. The way the mod works is to record the times where the boss HP, in %, changed. Since it is the only information available. The mod then 'graphs'/extrapolates an estimate of how fast the boss will die, but all of this assumes a constant DPS. It totally fails a taking into account simple things like bloodlust, or execute range. The mod is somewhat accurate for 100-20% boss HP, but past that point its not much reliable. And even so, you can't either rely on the timings from the start of time to die (ie: 50% boss hp+) since those data points also ignore the same things (bloodlust/execute range). This means at 50% it might tell you that you have 3:30 left, but in practice you have 2:58 left. Too bad you used icy veins so soon.
I don't know how many mages use that mod, but for the reasons it fails, the mod MobHealth succeeds. MobHealth will show a rough estimate, usually within about .25% of the actual health of a boss left, numerically and not percentage. When you pair the numbers from MobHealth, and the time left to Berserk on a boss mod, you get a good idea on how long you got left.
Bitterness is like cancer. It eats upon the host. But anger is like fire. It burns it all clean.
But the way mobhealth works is roughly the same ! Here is how mobhealth works.
The only data you know about any given mob, as given per wow itself, is the boss hp%. Mobhealth remembers how much damage is dealt to one mob to go from a percentage to another (ie: how much damage was dealt for the mob to go from 100% to 80%) then it extrapolates the total HP. If the boss heals, the numbers will get messed up. Also, the mod stores a database of previously calculated total_hp, so the numbers are more accurate. So now here is my point.
The mod in its current format is not able to display any mob health % smaller than flat percentages. ie: you will not see 1.3%. The only thing you will see is either 4% or 3%, because thats the only value available from the client. If the mod supported to dynamically display mob health% taking into account currently dealt damage from combatlogs, then yes, that would indeed work and be surprizingly accurate, but the fact remains that in its current state, it does not.
<Eej> YOU"RE GONNA PULL
<Eej> IF YOU SQUEEZE OFF ANOTHER ARCANE BLAST
<Spectear> You've obviously never played with Manly.
<Spectear> That's hardly a reason to stop DPS. Very Manly Staff
Hey - Roughly 50 pages back or so there was some speculation on new cookie-cutter/optimal gear setups, although some new items are yet to be discovered, I'd like your input on this one. (Oh well, the only one included here not "discovered" yet is Sunflare, but hey, it's there!)
Total Stamina: 421
Total Intellect: 370
Total Spirit: 107
Total Damage: 1387 fire
Total Critical rating: 297 (13.4%)
Total Hit rating: 160 (12.60% hit, 3% from talents, 1% from Shaman aura(Alliance), total 99% hit (capped))
Total Haste rating: 338 (Fireball: 2.38s cast - With permanent [Drums of Battle]: 2.26s)
04/04/08, 6:27 PM
Similar Threads
