Avoidance in MoP

even tho my basic strategy is hit>Exp (hard-cap)>haste>mastery>> parry > dodge

I noticed that I had trouble when there where lots of hard hitting adds to tank (like shek'zeer P2 in example)
I guess this is because we don't have one swing timer to worry about but lots of them, making a good timing for SoTR impossible.

While using Holy avenger for it was amazing, as soon as the SoTR buff fell off after HA and GoAK was over too, my healers where close to a hard attack.

so I tried around and went for more avoidance (through trinkets to keep it cheap) for this fight and the damage was incredibly smooth for the add phase.

so my question is, is there a proof (in a mathy way) that this is not just "a feeling" but that with more swing timers running and hitting you avoid becomes better in the way of damage smoothing? (I know that avoid gives you less TDR, I mean less spikes with more damage sources)

It does make sense in a mathy way. Consider what theck stated back when he was analysing the stats for MoP:

The "control" strategy is weaker in TDR, but grants better control on the 5-10 sec window. The reason behind it is mostly law of averages - 5-10 represents 2-5 swings from a boss, so one unmitigated hit is immediatly an extra 25 (going from 4 to 5 hits) to 100% (going from 1 to 2 hits) damage.

Adding multiple mobs is roughly equivalent to accelerating time while reducing swing damage. In a longer time period, control loses its edge to pure TDR.

I'm guessing theck will drop a more serious explanation, but the napkin math is: more weaker hits favor dodge/parry, as there is less chance for a long string of direct hits.

The phase is pretty hairy though, if your doing the 2/6 split and tanking the 6 smaller adds, it is painful even with more dodge and parry, basically it's down to the DPS to kill 1/2 of them before your CD's are gone, they have around 30 seconds (GotK, AD, HA + DP) if you spread your cd's out to finish at least 2, I find after that tanking them gets pretty trivial. Even if you are going 4/4, the dps need to focus 1 ASAP (I found the 4/4 split to be fairly easy to tank after just 1 add died, even thought the bigger adds hit a bit harder).

Some minor kitting might be a decent idea, should they not die in time, frontalobe had the excellent idea of glyphing Blinding light for the knockdown, that gives you some kiting breathing room or a further 2 seconds of limited damage.

It's a rather isolated fight there isn't really another fight like it in this teir, I still maintain the last thing you want is to be more spikey in that fight than you already are.

Schroom wrote:so my question is, is there a proof (in a mathy way) that this is not just "a feeling" but that with more swing timers running and hitting you avoid becomes better in the way of damage smoothing? (I know that avoid gives you less TDR, I mean less spikes with more damage sources)

I don't think more avoidance would help smooth the damage, with 5+ mobs hitting damage intake should be quite smooth. In fact adding more avoidance would make the damage intake more spikey. However these spikes would just be the same as the smooth damage you would take not wearing avoidance gear so it is preferable.

Schroom wrote:so my question is, is there a proof (in a mathy way) that this is not just "a feeling" but that with more swing timers running and hitting you avoid becomes better in the way of damage smoothing? (I know that avoid gives you less TDR, I mean less spikes with more damage sources)

You have it backwards. Avoidance is generally better for TDR, at least in the steady state. The advantage of control/haste is that it smooths out damage spikes over several hits. With avoidance, there's always a chance you'll take 3-4 unavoided attacks in a row, because avoidance is random. With control, you can guarantee that some of those attacks are covered by SotR.

That's assuming a 1.5-second swing timer though (and longer generally makes control even better). What if the swing timer is shorter? Well, the shorter the swing timer, the closer you are to a true stochastic situation, by which I mean that you can start to rely on the averaging of avoidance a little more. The chance of avoiding nothing still exists, but it gets smaller and smaller the more hits you consider. The chance of avoiding 0 hits out of 3 is much larger than 0 out of 6, for example. The first is (1-avoid)^3, the second is (1-avoid)^6.

To see what practical effect that has on smoothing, we can use the same simulation code I used here to look at damage smoothing for a boss with a 1.5-second swing timer, but reduce the swing timer to ~0.25 seconds to simulate having 6 different attackers (each with a 1.5-second swing timer). Since we're being attacked 6x faster, we'll want to consider longer strings of attacks. 15 attacks is about 3.75 seconds of combat, and 30 is 7.5 seconds, so those seem like reasonable bounds.

Here's what the data looks like:

Code: Select all

15 attacks:

| Set: |     #1 |     #2 |     #3 |     #4 |     #5 |     #6 |     #7 |     #8 |     #9 |
|   S% | 0.3768 | 0.4156 | 0.4246 | 0.3470 | 0.3964 | 0.4526 | 0.4516 | 0.4143 | 0.4147 |
| mean | 0.5816 | 0.5988 | 0.6042 | 0.5603 | 0.5900 | 0.6206 | 0.6212 | 0.6016 | 0.6038 |
|  std | 0.1364 | 0.1326 | 0.1294 | 0.1359 | 0.1396 | 0.1195 | 0.1189 | 0.1315 | 0.1206 |
|  80% | 6.5699 | 8.0943 | 8.1326 | 4.4090 | 8.0906 | 8.0134 | 7.8593 | 8.0838 | 6.0530 |
|  90% | 0.6082 | 0.9031 | 0.9252 | 0.3469 | 0.8179 | 1.0351 | 1.0105 | 0.8930 | 0.6979 |

Code: Select all

20 attacks:

| Set: |     #1 |     #2 |     #3 |     #4 |     #5 |     #6 |     #7 |     #8 |     #9 |
|   S% | 0.3768 | 0.4156 | 0.4246 | 0.3470 | 0.3964 | 0.4526 | 0.4516 | 0.4143 | 0.4147 |
| mean | 0.5816 | 0.5988 | 0.6042 | 0.5603 | 0.5900 | 0.6206 | 0.6212 | 0.6016 | 0.6038 |
|  std | 0.1096 | 0.1028 | 0.1003 | 0.1123 | 0.1087 | 0.0929 | 0.0911 | 0.1011 | 0.0969 |
|  80% | 3.2809 | 3.7318 | 3.7822 | 2.1616 | 3.8650 | 3.6538 | 3.2228 | 3.5541 | 2.7566 |
|  90% | 0.1943 | 0.2737 | 0.2566 | 0.1086 | 0.2303 | 0.2644 | 0.1750 | 0.1948 | 0.1510 |

Code: Select all

25 attacks:

| Set: |     #1 |     #2 |     #3 |     #4 |     #5 |     #6 |     #7 |     #8 |     #9 |
|   S% | 0.3768 | 0.4156 | 0.4246 | 0.3470 | 0.3964 | 0.4526 | 0.4516 | 0.4143 | 0.4147 |
| mean | 0.5816 | 0.5988 | 0.6042 | 0.5603 | 0.5900 | 0.6206 | 0.6212 | 0.6016 | 0.6038 |
|  std | 0.0900 | 0.0823 | 0.0806 | 0.0950 | 0.0862 | 0.0764 | 0.0734 | 0.0791 | 0.0805 |
|  80% | 1.5348 | 1.5153 | 1.4915 | 1.0050 | 1.6082 | 1.4785 | 0.8559 | 1.0007 | 0.8964 |
|  90% | 0.0498 | 0.0668 | 0.0551 | 0.0282 | 0.0498 | 0.0641 | 0.0053 | 0.0160 | 0.0151 |

Code: Select all

30 attacks:

| Set: |     #1 |     #2 |     #3 |     #4 |     #5 |     #6 |     #7 |     #8 |     #9 |
|   S% | 0.3768 | 0.4156 | 0.4246 | 0.3470 | 0.3964 | 0.4526 | 0.4516 | 0.4143 | 0.4147 |
| mean | 0.5816 | 0.5988 | 0.6042 | 0.5603 | 0.5900 | 0.6206 | 0.6212 | 0.6016 | 0.6038 |
|  std | 0.0778 | 0.0714 | 0.0707 | 0.0833 | 0.0732 | 0.0688 | 0.0654 | 0.0672 | 0.0708 |
|  80% | 0.6793 | 0.5848 | 0.5565 | 0.4563 | 0.5801 | 0.5529 | 0.1090 | 0.1368 | 0.1929 |
|  90% | 0.0096 | 0.0157 | 0.0099 | 0.0060 | 0.0078 | 0.0125 | 0.0009 | 0.0003 | 0.0003 |

The gear sets are described in the blog post, but in short:
#1: ~4.5% hit/exp, moderate avoidance, high mastery, no haste
#2: ~7.5% hit/exp, low avoidance, high mastery, ~4.5% haste
#3: ~7.5% hit/exp, low avoidance, high mastery, ~7% haste
#4: low hit/exp, high avoidance, high mastery, no haste ("avoidance build")
#5: ~7.5% hit/exp, low avoidance, very high mastery, no haste
#6: ~7.5% hit/exp, low avoidance, low mastery, ~14% haste
#7: ~7.5% hit, ~15% exp, low avoidance, low mastery, ~8.7% haste ("control/haste")
#8: ~7.5% hit, ~15% exp, low avoidance, high mastery, no haste ("control/mastery")
#9: ~7.5% hit, ~15% exp, high avoidance, very low mastery, no haste ("control/avoidance")

The data suggests that over short periods (~4 seconds, or 15 hits), the hit/exp/mastery sets aren't as good as the high-avoidance sets at reducing spikes. #4 stands out here, and it uses a combination of high avoidance and high mastery. #9 isn't that far behind though, which is a hit/exp-capped set that focuses on avoidance.

This makes a lot of sense, too: you simply can't make sure that SotR is up to cover a part of every 4-second period with normal HP generation, so the "control" setups are going to have spike periods when SotR is down. The stochastic nature of avoidance is making it better over these time scales, because you're taking lots of small attacks and avoiding some of them. That makes the avoidance build ~2x as good at removing spikes above 80% of max possible damage intake, and ~3x as good at removing spikes above 90%.

As we move to longer time scales, the control sets (#7 and #8) start to pull ahead. Over a 5-second period (20 attacks), the gap narrows considerably, and by 25 attacks control/haste has pulled solidly ahead in all categories, and beats avoidance by a factor of 10 in spikes above 90%.

So it depends on what you're trying to optimize. Avoidance will give smoother overall intake for periods of <4 seconds, control/haste will generally win at 5+ seconds. Note that this also ignores cooldowns, which would tend to favor control strategies as you can optimize your SotR coverage by banking HP during cooldown effects and burning it between cooldowns (i.e., you could alternate cooldown-SotR-SotR-cooldown-SotR-SotR-, so that you have 6 seconds of SotR uptime in-between other DR cooldowns to stretch out your coverage window).

yeah I thought as much. so I chose for that fight I stayed at 7,5% hit 15% Exp and ~8% Haste and simply switched the shado-pan trinket that gives me stamina and mastery "on use" (I "used" the trinket as a supplementary Cooldown in my CD-rotation to tank those adds the night before) with the LFR version of the Elegon trinket that gives me some mastery (less than the use of the shadopan one) forged to parry and a dodge procc.

so in essence I switched mastery for avoidance.

which would be #7 before and somewhere around #9 after

the phase is pretty long so in fact my choice and my "Feeling" where correct I guess. Also my healers went out of it with a much lower pulse

Isn't the point that with lots of little attacks (from adds etc), judging gear by a TDR metric becomes more sensible than judging them based on ability to smooth damage? They are little attacks, so even if you get bad luck with them, it does not add up to a big spike and they are many, so the damage they generate is smoother anyway. From what Schroom was saying it sounded like a healer throughput issue - he was taking so much damage, they were sweating to heal him. With avoidance, he took less damage and the healers were happier.

Now it may be that the account is just subjective - Schroom spoke of it being a "feeling" and wanting maths to back it up. One of the biggest issues with avoidance seems to be that diminishing returns mean that even if you gear for it, it does not dramatically affect how much you avoid. If I am reading Theck's tables right, gearing for avoidance only increases your average damage reduction by 2 points, from 0.58 to 0.60, which does not seem worth getting excited about and should not really be perceptible by your healers.

One question I have with avoidance in MoP is trying to quantify the trade offs. Under the "control" weightings of MrRobot, avoidance is IIRC 0.25 of the weight of either mastery or haste. This a much bigger difference than I had expected from a casual reading of some of Theck's work. I had been taking away the message that avoidance stats are good tank stats (they are best for reducing average damage), it's just that but mastery/haste/hit/expertise are (a little) better (because they smooth damage and that's more important for tank survival). But with those weights, it's more like avoidance is a dump stat of very little value. ("/spit avoidance" was how one person on the MMO Champion paladin forum put it).

For example, the weights are likely to imply a dps piece with crit will be better than one with double avoidance. For example, AskMrRobot is telling me to buy the new 496 VP boots with expertise/crit instead of those with dodge/parry. I accept the logic of the control argument and looking at what the extra dodge/parry would actually do to my avoidance, I can see it's not much (imperceptible, to be frank). However, there is a residual doubt about spending my hard earned VPs on something with crit (I am very slow in raiding progression, so don't tend to think much about drops: according to MrRobot, the boots are by far the biggest VP upgrade open to me atm).

Schroom wrote:yeah I thought as much. so I chose for that fight I stayed at 7,5% hit 15% Exp and ~8% Haste and simply switched the shado-pan trinket that gives me stamina and mastery "on use" (I "used" the trinket as a supplementary Cooldown in my CD-rotation to tank those adds the night before) with the LFR version of the Elegon trinket that gives me some mastery (less than the use of the shadopan one) forged to parry and a dodge procc.

so in essence I switched mastery for avoidance.

which would be #7 before and somewhere around #9 after

the phase is pretty long so in fact my choice and my "Feeling" where correct I guess. Also my healers went out of it with a much lower pulse

I'm not sure it's worth swapping at all. You should have more than enough cooldowns to chain to cover the first minute or so of tanking the adds, and after that point several should be either dead or chasing other players, so the throughput drops dramatically. In essence, you're already covering the part of the encounter where 3-4 second spikes are dangerous with cooldowns, so it's not really advantageous to gear for those spikes. The part you can't cover with cooldowns is low-throughput, meaning it'll be 5+ (often longer) second spikes, which control dominates for (and your healers should be able to heal through, in any event).

Unless you're not chaining cooldowns, in which case I'd call that a strategy fail, not a gearing fail.

econ21 wrote:Isn't the point that with lots of little attacks (from adds etc), judging gear by a TDR metric becomes more sensible than judging them based on ability to smooth damage? They are little attacks, so even if you get bad luck with them, it does not add up to a big spike and they are many, so the damage they generate is smoother anyway.

Not necessarily. Lots of small attacks that come quickly can still be dangerous. It's spike DPS that matters. Getting hit once every second for 500k is going to be essentially the same as getting hit 5 times a second for 100k in terms of spike danger. You're still taking 1000k damage in two seconds, and your healers only have about half of that time in which to react and heal you.

The real difference in avoidance is that you're far less likely to take 10 unmitigated hits in a row than two. When you have a high enough attack rate, you edge closer to the stochastic situation in which avoidance shines.

But as we see in these simulations, what really matters in the high-attack-speed situation is the duration of the spike window. When the time window is short compared to the SotR duration, avoidance will start to pull ahead of control (provided, of course, it's unpredictable spike - predictable spikes highly favor control for obvious reasons). When the time window gets long enough that you can guarantee partial SotR coverage of the spike, control tends to dominate.

We saw the same sorts of results in the earlier bog posts for a 1.5-second swing timer, with the primary difference being that avoidance didn't pull ahead nearly as much for low swing rates. That's again due to basic statistics - avoidance performs much better if it gets 6 chances to work every 1.5 seconds than it does if it only gets 1 chance.

econ21 wrote: If I am reading Theck's tables right, gearing for avoidance only increases your average damage reduction by 2 points, from 0.58 to 0.60, which does not seem worth getting excited about and should not really be perceptible by your healers.

That's percentage of maximum damage intake. So you take 60% of the max possible (which is the boss hitting you with every attack, no mitigation) with one set, and 58% with the other. That's a net damage reduction of about 3.33% going from the control set to the avoidance set. Nothing to sneeze at, but not likely to be enough to change un-healable throughput to healable throughput. I doubt a healer would even notice that difference given that 30% of most healing is overheal to begin with.


Again, I'd remind people to take any anecdotal data with a grain of salt. A healer saying "gee you were so much easier to heal" is basically worthless, because it's not a controlled environment. Maybe you got lucky and took fewer spikes this attempt than you did in the other gear set. Maybe there was some extra cross-healing by another healer that they weren't aware of. Maybe you juggled your cooldowns better or more effectively this time. Maybe the healer was just more comfortable with the encounter this week than last, or managed their mana better in an earlier phase, etc. Or maybe they're full of shit and it's all in their head - it's easy to convince yourself of something that isn't true when you're not looking at tangible results.

I agree with Theck, control build seems much better for Empress on 10 man normal and I'd guess on heroic as well.
Another advantage to a haste build is more dps so the adds die even faster, on our last kill I did 30% of the total dmg to the windblades.

Maybe you got lucky and took fewer spikes this attempt than you did in the other gear set.

or maybe I had just bad luck in 1 or 2 tries last time.

yeah that's the reason why I wanted it backed up with math

My CD roation was the same, I got a chain system once and there it is so no problemo. The healer who was most responsible for me in this Phase was our druid, our holy priest watched over our warrior while the disc priest mostly group an crosshealed the tanks.

and basicaly I died as soon as GoAK and the monster SoTR buff from HA fell off down to glyphed DP. >50% reduce (with 4 piece bonus) down to 20% can be shock to the healer I guess if not prepared, some bad luck with avoidance and there it is.

also I noticed that the adds are level 92 which gives us 4.5% more avoidance by default against them.

Another advantage to a haste build is more dps

yeah my basic build is , as I said a hit>Exp>haste build and it was enough for a world ranking on that kill.

btw here is the video if someone is interested: http://www.youtube.com/watch?v=AOCeChTzjqI (hope you don't mind the german.)

so if there are any suggestions to my play stile feel free ^^ I can't improve without critics.

#offtopic

not so much about avoidance but about how you can take out a lot of dmg in this fight

i'm not sure if what i've seen is correct but are you tanking the big and the small together?

We are 25 man guild but allways splitting up the adds so 1 tank (dk) is tanking the 2 big ones and the other tank (me) is taking the small adds. With the small you have a little burst window which can easyly covered by your own cooldowns and another point is that i'm tanking the small ones behind the raid

---------- Stairs --Y-------
addspawn -Z-----Z-- addspawn
----------- Raid -----------
------------ X -------------

X is the position where i usualy tank all the small adds.
Y is the position where the dk is tanking the 2 big ones

I don't know how much the mechanic in 25 is different to 10 but i saw a trap in your video. (had to interupt it as boss is sneaking through the office atm XD ) . If you don't tank the big and the small together the chance of getting the big one (which is a huge advantage) is of course lower as if you just move both big adds over a trap.


But with this the small adds can be aoe stunned, bombed and at some point the will start to fixate raidmembers anyway

As soon as you can place a trap at the left Z the dk is moving his adds over it and 1 gets trapped -> all the small adds are jumping onto the big add. I will fraps it next time we will kill it.


after reading my post for typos and stuff -:> i should have written it in german, so at least you could understand what i mean XD

I get it. it is a fair enough idea. the only tricky part should be the start of P2 for both tanks to split the adds.

we just took them as they came. our warrior went on the left side, while I took the right one.
usually I need less heals and get less spikes as I have a big gear advantage over our warrior. so our DPS first killed as much small adds from the warrior as possible, building one trap, the warrior leaping behind it, kiting his bis add into the trap, then taunting my big add cleanse, repeat, next phase. in the video we did the achievement killing both big adds not more than 10 seconds apart. that's why the warrior put a small add into the first trap. and me putting my big add into the second trap.

The idea expressed in the OP is one that has always been traditionally accepted in principle in WoW tanking, even independently of paladin-specific concerns.

In order to rigorously test the degree to which you would practically benefit from that gearing strategy obviously we have to do a concrete analysis like Theck's, but intuitively it is absolutely sensible that our subjective perception of what avoidance is doing changes as a function of frequency and size of individual swings.

In the mathematically-limiting case where rate of swing increases without bound while swing size approches zero, avoidance would be empirically indistinguishable from armor as far as the characteristics of the effect, at least as far as our perception over non-microscopic timescales.