Ok thanks, after reading the spriest threads and some other information i could find the biggest things i can see to improve her dps are dot uptimes and spell priorities. I guess that's probably the answer to improve any spriest dps but yeah...
Any other suggestions apart from those.
Added a Ignis WWS to my original post incase that's easy to critique.
Added a Ignis WWS to my original post incase that's easy to critique.
Just from looking at it i can see a few things that could help her improve, Ignis is a pretty hard fight to keep a decent dot alocation up due to when flame jets comes it can sometimes be 2-3seconds before you VT might go up for example however there is no reason why she couldnt hit 4-4.5k DPS.
1. She is under hit cap, this probably isnt a real issue as its only 10.1/11% (290 ish), i can hardly see this making the difference but
2. The fight lasted 272 seconds, while she only did 24 mind blasts. She could have done about 35-40 easy (272/6 = 45)
3. Dot time up, They are all averaging at about 3:50ish. (272/60 = 4.5mins) this is an ok time i guess, i would personally be aiming for at least 85% depend on how many slag pots i got.
Suggestions -
1. Ask if she is using a mod like Quartz, this allows you should a dot timer for your target, i personally find this extremely handly and personally think its make things a whole lot easier, this being said however, she can easily just make her debuffs on the boss bigger or something to help keep track of it.
2. Make sure SW:P is getting applied after there are 5 stacks of misery, although its not affected by SPower, this does affect it. Nice opening i find is VT>DT>MB>MF full duration>SW:P, this will mean that she will not get threat for about 3seconds and lets teh tank pull in to position, Great on a fight like XT. VE can also be thrown in since its pretty much a must on most Ulduar bosses.
3. Remind her of priorities, VT>MB>DT>MF. that being said however, SW:P must never fall off. Also i try not to use SW as much in Ulduar, just puts more pressure on the healers. Save it for when you get jetted in ignis or something for example, which from the WWS it looks like she is doing that.
In an effort to increase my dps as shadow, I have spent some quality time on the training dummies doing sustained dps checks. I pull off 3k dps, sometimes upwards of 3.2k dps. My DOT uptimes are typically 82-85% VT, 90% + DP compared to the shadow word pain. I have resorted to using Mind Flay as filler, often times interrupting it after 1 or two ticks to keep the VT>DP>MB cast priority. After reading a post above on the VT mechanic regarding clipping the last tick, my dps has increased quite a bit, and I am able to keep my dps up at what I get from the buttom smashing maro method (earlier post, refering to a post on shadowpriest.com). What do some of you real sahdow priests get on the target dummy?
My issue appears to be mind blast, using calculations off recount for the total time / MB casts, I get around a MB cast per 8s, best being 7.8s per MB cast. I was doing raid boss target dummy for 450-560ish seconds last night. It just seems that I should do better with MB, I have quartz cast bar, DOT Timers and Power Auras to help me manage the rotation, and I swear I casted MB as the cooldown timer on the MB icon just ended or in worst case, within a GCD. is there somethign I am missing? It shouldnt be that hard to cast MB every 6.5-7s!
As for dps, i'd like it to be nearer 3.5k on target dummy, is it gear? I accumulated a few peices of dps gear, and combined with my heal set, I have OK stats, though crit is lacking. Hit capped, I adjsuted peices to take hit to minimum 290, and I didnt see any misses on Recount, so I should be good in that regard.
Using only self buffs + 1x shadowfiend, I average 3400-3500dps on a target dummy. I think you shouldn't worry about 3000 dps, seems fine to me considering your gear. Just one little thing, instead of using only 1 tick of mindflay, try to use SW:D instead - maybe it will raise your dps.
Is RAWR completely inaccurate when judging your possible (theoretical) dps? I've been looking around and trying to figure out why my Patchwerk parse is so below what RAWR suggests for a similar-length fight with the same buffs and priority system.
Patchwerk Parse: WoW Meter Online - Combatlog Replay
RAWR suggests 5704 DPS over a 4 minute fight with the same buffs and debuffs available.
Also, any suggestions as to upping my DPS would be amazing. As it stands on the Patchwerk parse, I 'missed' (forgot/ignored) a DP refresh, but I doubt that 85% uptime vs 90+% would cost me nearly 1k dps. I landed 30 out of 33 possible mind blasts, and my Shadow Word: Pain and Vampiric Touch uptimes were above 90% (95.7% and 91.8% respectively), unless I'm doing the math completely wrong.
From my comparison of Rawr vs Parses Rawr seems to consistantly overshoot DPS by atleast 10%.
I'm wondering if you might be able to explain how the "Game/Brain Latency" operation works within Rawr (V2.2.10), since you code the priest portion of it. I have a suggestion about its usage, but my assumption of how it operates could be wrong. I'm curious to know how you and other players use this function as well, so any anecdotal evidence would help (ie never touch it, set it low, set it at xxx, etc.)
The reason I'm asking is because Rawr was consistently over calculating DPS compared to my own performance. Additionally, I've noticed others who report the same phenomena in this thread. While I recognize that Rawr isn't a perfect model AND that human performance falls short of perfection, I think the overestimation is a little too high and that perhaps the "Game/Brain Latency" option may not be correctly used by most players.
What drew my attention to the "Game/Brain Latency" function is that my job has me working with Human Factors experts and dealing with Human Reaction times in a variety of situations. In the context of WoW and damage rotations, our reactions are considered Complex Reaction types. This is because we have to Detect a stimulus, Recognize the stimulus within the brain, Decide on our response, then Perform the response. The time it takes to perform these actions varies considerably on the situation, however some estimates are provided. For example, signal Detection is estimated at .1s, Recognition .4s, and Decision time up to 4.0s. All of this would have to occur before the body begins to move.
Now, we obviously don't take 4 or more seconds to make decisions when DPSing. Things such as practice (playing often) and pre-cuing (watching DoT timers) help us to cut down considerably on the decision making time and motor response time. However, we're still limited by the amount of time it takes to perceive that we need to begin casting a spell, and actually doing it.
I believe it is reasonable to conclude that we often times have more than 2 or 3 choices to make during boss fights, which includes things other than straight DPSing (ie talking, moving, etc.) For tasks that contain just 1 choice, the average reaction time is .2s; for 2 choices, .35s, and for 3 choices, .4s. Therefore, after reviewing different reaction time tables and fiddling with the "Game/Brain Latency" myself, I would suggest that players add 300-400 ms to their game latency, and input that number into Rawr before determine their DPS.
In practice, I found that the difference in DPS between 0 ms and 400 ms on my character was ~450 DPS (I tested this without using any buffs except self buffs, as if I was attacking a Target Dummy. Time was limited to 1 minute, to remove any mana limitation). This was a 16% decrease in DPS for my character based on Rawr's numbers. If I took my game latency into account (about 100ms), then the difference between 100ms and 400ms was about a 12% decrease in DPS. The end result of all this was that my performance on a Target Dummy in game as compared to Rawr’s DPS numbers was much closer. Though Rawr was still higher in its estimation, this was expected do to personal performance variations.
As you noted, Rawr numbers seem be off about 10% based on WWS reports. It may be that players are not be setting their "Game/Brain Latency" high enough to receive realistic numbers. It might be worth mentioning to other players that using this setting correctly may help them get a better estimate of where their actual performance lies compared to a realistic-theoretical performance.
Resource:
1. Woodson, W.E., Tillman, B, & Tillman, P. (1992). Human factors design handbook. New York: McGraw-Hill.
You nailed it. The idea of the Net Latency/Brain Latency is to take into consideration both things you cannot control (Network Latency), things you can control (movement/decision) and finally things you cannot control as well (finger slipping and similar failures).
I did talk to my guildies and they agreed that my default of 100ms is "figther pilot" values, and that a more realistic value is perhaps closer to 300ms. You are suggesting values as high as Ingame Latency + 300-400ms. Unfortunately, I never REALLY looked into how much this lag adds up to and left it at 100ms.
To comment on the rest, I don't think a seasoned WoW player has up to 4.0s decision time, also the decision time slightly overlaps since we can prepare for what is going to happen before we have to do it. There is a pipeline of which you are placing commands ready at your fingertips before they need to happen, this eats up a lot of the Signal Detection, Recognition and Decision time, without causing any delay. Of course if you suddenly have to move out of fire or void zones or just simply going out of range, you are going to have to get penalized with the entire process.
I should possibly try to make people more aware of this option + tweak the default value to something higher.
Rawr - Coder of HolyPriest (Healer) and ShadowPriest (DPS) Modules. Get Your Rawr 2.3.x!
Yes, 100ms is a "fighter pilot" value, which comes from both practice and highly efficient ergonomics. That is, they operate within a framework that is fine turned for human operation. Everything from text size, display luminosity, distance between buttons/controls, to name a few, all come into play.
One thing I didn't mention, and part of the reason why I chose a "game latency + 300-400ms" value is because of eye movement. Depending on how far your eye moves away from your DoT timer, for whatever reason, it can take 100-200ms to get back (this includes everything; synapses firing, focusing, etc.) Even though eye movement doesn't factor into every single reaction when casting, it still makes an impact over the course of a fight. Because the computer screen typically sits within the optimal visual range for both eyes (15 degrees left and right of center), and since head movement is limited or non-existent, I felt that .1s was a reasonable range to consider, hence the 300-400ms.
While this range may seem like a lot to some, its actually a very conservative estimate on my part. I think that game latency + 450ms is more realistic for the average WoW player. Conversely, I wouldn't be opposed to a game latency + 250ms value for WoW veterans.
I've done some small studies on reaction times. The summary is that the 100ms is the absolute fastest reaction you'll get to an UNEXPECTED event, and the slowest reactions are around 350 ms. 120 ms is considered "really good" and 200 ms to 250 ms is average.
Keep in mind, however, that this is reacting to an unexpected event. For example, this would include being randomly targeted for a shadow crash on Vezax. It is not an appropriate estimate of reaction time to an expected event, such as the exact moment in time your Mind Flay will finish channeling. I haven't tested reaction times to expected events, but I suspect they are on the order of 20 ms to 100 ms for people who are really concentrating on things.
I've done some small studies on reaction times. The summary is that the 100ms is the absolute fastest reaction you'll get to an UNEXPECTED event, and the slowest reactions are around 350 ms. 120 ms is considered "really good" and 200 ms to 250 ms is average.
Keep in mind, however, that this is reacting to an unexpected event. For example, this would include being randomly targeted for a shadow crash on Vezax. It is not an appropriate estimate of reaction time to an expected event, such as the exact moment in time your Mind Flay will finish channeling. I haven't tested reaction times to expected events, but I suspect they are on the order of 20 ms to 100 ms for people who are really concentrating on things.
I'd be interested in seeing the studies you looked at. And I'd also like to know how you view reaction time. That is, do you include the time to Detect, Perceive, Decide and Respond as all-inclusive of reaction time? The reason I bring this up is because the lay definition of reaction would be inclusive of all those things. It just makes sense right? But a lot of times, studies talking about reaction time are really only focusing on the time between when a signal is detected and the first response of the body. For example, if a spec of dust hits your eye, the eye closes. The dust hitting your eye is the stimulus signal. The signal to close the eye is the reaction. However, the actual time to close the eye is not considered. This is also a simple reaction, and involuntary. It skips decision making time, which is where majority of the total response time comes from.
When it comes to our actual performance in game, 120ms seems way too low, due to 1) multiple events being processed, 2) foveal vision comprising just 2 degrees of the visual field, requiring subsequent eye movements, and 3) motor response time of the hands/fingers.
While I realize anticipation can improve our reaction time, we really do have a lot of things going on during a dps phase. Even if we look at a Patchwerk type fight, where we can focus purely on DPS, you still experience the those 3 points mentioned above. If people really are anticipating their reactions such that they perform the necessary key strokes between 20-100ms, I would expect to see people coming within 3% of their theoretical DPS from Rawr. Perhaps people have, but haven't posted about it. When I get home from work later tonight, I'll try to run some Target Dummy tests. Whatever the result, the setting for the "Game/Brain Latency setting" will most likely need be greater than 250ms to get more realistic results. I still feel setting it to game latency + 250ms would hit close to the sweet spot for WoW veterans.
I ran some tests with my priest to see what Game/Brain Latency setting in Rawr would be the most likely candidate for producing DPS numbers that were close to actual performance seen in WWS reports. I use MFclip and a self made dot timer that watches only VT, DP and SW:P. (I did not edit my previous post, as my opinion changed).
Experiment
Before I logged into WoW, I loaded my priest from Armory into Rawr. I unchecked all buff boxes except for Inner Fire and Divine Spirit, as these add to DPS. I also removed both trinkets from the setup, so that their procs would not interfere with the data. After I logged into WoW, I removed all buffs on myself, except for Inner Fire and Divine Spirit to match the Rawr set up, as well as my trinkets. I then proceeded to a Level 60 Target Dummy to begin my tests. A Level 60 Target Dummy was used so that Hit % would not be a factor.
To check baseline data, Rwar Options were set to:
Relative Mob Level:
0
Mana Potion:
None
Time spent in FSR:
100%
Fight Length
2 Min.
Game/Brain Latency
99ms
Shadow Fiend
0
Replenishment
100%*
JoW
0
Survivability
0
Spell Priority was set to default, except Shadow Word: Death was removed from the rotation. It is a DPS loss for my current gear set up and would not be used in testing. The resulting priority list was: VE, VT, MB, DP, SWP, MF.
The resulting theoretical DPS-Burst was: 3127
The resulting theoretical DPS-Sustained was: 3078
With all other settings held constant, and Game/Brain Latency set to 350 ms (100ms latency + 250 ms reaction time):
The resulting theoretical DPS-Burst was: 2828 (9.6% difference)
The resulting theoretical DPS-Sustained was: 2779 (9.7% difference)
Results
I was able to run just 4 tests, so the data sample is admittedly small. Tests 1-4 yielded:
Exp. #
Dmg. Per. Sec.
Time (s)
Total Damage
DPS, if T = 120s.
1.
2726.1
115.75
315,546
2629.6
2.
2773.2
115.66
320,748
2672.9
3.
2675.7
115.42
308,829
2573.6
4.
2708.8
114.69
310,678
2589.0
Avg.
2720.9
115.38
313,950.25
2616.275
Discussion
When considering the true test times, the resulting average DPS was approximately 11.6% short of the Theoretical DPS-Sustained @ 99ms, and 2.1% short of the Theoretical DPS-Sustained @ 350ms. When considering the 2 minute test time, the average DPS was approximately 15.0% short of the Theoretical DPS-Sustained @ 99ms, and 5.9% short of the Theoretical DPS-Sustained @ 350ms. It is important to note that my DPS is clearly lower when we look at a time length of 120 seconds, instead of ~115 seconds. The reason my tests weren’t a full 2 minutes is that my gear cannot support a full DPS burn for more than ~115 seconds at this point. Therefore, adjusting my DPS to match a full 2 minute combat length is more appropriate for evaluating Rawr and the Game/Brain Latency setting, since Rawr’s theoretical DPS values were based on a full 2 minute fight. Though negligible, there will be some error, since Replenishment causes some mana regeneration during those 5 seconds, possibly allowing 1-2 more spell casts.
Ultimately, the experiment shows that Rawr DPS estimations are overestimated by at least 10%; in this case ~11-15%. While this is a single case experiment limited by my own performance and the data set was small, I think it at least gives some credence to the notion player performance in game will be roughly 10% short of Rawr estimates when the Game/Brain Latency setting is left at the 100ms default. The experiment also shows that adjusting the setting to “game latency + 250ms reaction time” brings performance DPS within about 6% of the theoretical DPS. While some DPS loss is likely due to human error in the priority rotation (such as casting a MF longer than needed, or casting MB when VT needs a refresh), it isn't likely to cause a 6% DPS loss. Therefore, the majority of that loss can be attributed to timing and reaction time.
After conducting these tests and, in particular, noting the benefits in reaction time that the MFclip addon provides, I feel that my previous statements about reaction time were judged too high. Because the MFclip addon helps us combat the latency component of the video game and because of the way it cues a player visually, I think that even average, less reactive players will still be able to perform in the 200-300ms range, not 300-400ms. Therefore, I would suggest that most shadow priest using Rawr set the Game/Brain Latency setting to “game latency + 250ms”. Even if players don’t perform at that level, I think it would allow for a realistic goal for players to achieve. Additionally, an experienced player might find that “game latency + 200ms” a more realistic performance goal.
*Actual replenishment time was ~99%, due to VT having to be up before Mind Blast. This did not make a difference in Rawr DPS calculations.
It seems as if you're casting VE in rawr but not in your tests.
Having VE in the rotation doesn't effect Rawr's DPS output. Even if it did, it would only be 1 GCD, and would constitute a near undiscernible error in the experiment.
It doesn't affect rawr's dps? That's a bit strange. It's one GCD over the course of 120 seconds, so it's more than 1%. Whether that's indiscernable or not is subjective, but you may as well take it out since it doesn't add any benefits.
The Not So Evil, if I could ask you a technical question, how exactly does RAWR model latency?
Is it single value (with possible random element) that increments the variable controlling when the next action can be cast or something to that effect?
Or does it try to model the client/server model that is used in game in some way and thus with variable latency etc. can lead to clipped ticks or DoTs and the like?
It is done as simple as possible. (So its what you described first). Clipping ticks or DoT's is not modelled as thats up to the player to avoid. I could add such options, but feel it would just confuse most people when it doesn't work as they expect.
Rawr - Coder of HolyPriest (Healer) and ShadowPriest (DPS) Modules. Get Your Rawr 2.3.x!
It is done as simple as possible. (So its what you described first). Clipping ticks or DoT's is not modelled as thats up to the player to avoid. I could add such options, but feel it would just confuse most people when it doesn't work as they expect.
Yeah that's pretty much the approach we use in SimulationCraft as well, but I'm curious as to what difference show up with variable latency and the like if we did model it "properly".
I'm Oceanic, so variable latency and higher latency is something I experience every raid.
I think it would be interesting to determine the difference between the various classes when such things need to be accounted for.
To be more accurate regarding how we currently handle it in SimulationCraft, we maintain 3 different latencies values with the correct one chosen (using gaussian distribution RNG with separate variance values for each type of latency) depending upon if the last spell was a normal "queued" spell (say, Mind Blast or Frost Bolt), a GCD instant cast spell (SW:P), or a channeled spell (Mind Flay). (default values are 75ms for queued, 150ms for GCD and 250ms for channeled).
This RNG'd latency value is just added onto the effective cast time of the previous spell thus pushing back when the next spell can be cast.
When it comes to queued spells unless you really do have shockingly high or highly variable latency then you're really only limited by how often you hammer your buttons near the end of the cast time. If you're a robot (or you're using a G15 or AutoHotKey) you can pretty much reduce this latency down to 0ms no matter what your fixed connection latency is or any brain-lag, so a default of 75ms seems okay if you're not perfect at button spamming.
To be quite honest, I'm not so sure that GCD spamming suffers any more than queued spell spamming (would need to test this again...tested it a while back but forgotten the results) if you're button mashing, but handy to have the option to model it differently anyway I guess.
channel_latency is there of course because with channeled spells you're given the option of using [nochanneling] macros where your fixed connection latency actually does matter to your rotation as it will directly increase the channeling time (note, it doesn't for GCD and Queued spells) or you rely upon trying to "stopcast" early predicting the connection latency and thus possibly running afoul of clipping. While SimCraft has the code to support clipping of DoTs etc. it's not something that is used with connection to latency and the current model basically has the effect of using lots of [nochanneling] macros and treating channeled spells like queued-spells but with connection latency added in.
Thus my interest in changing to a more "realistic" model.
(I'm the SimulationCraft Priest module maintainer obviously)
*edit* Obviously, I'm talking about a Patchwerk/Test Dummy style fight where as a Shadow Priest you don't have to react to the environment and can just focus on your rotation/priority list.
I fully agree that when it comes reacting to events like Shadow Crashes, Void Zones and Storm Clouds as well as ones like Heroism about to be popped/being popped (Is there time to pop Shadowfiend or a Wild Magic pot etc.) that you'd need to factor in "brainlag" into things as well as it could quite definitely have the chance (depending upon where you are in your current spell cast) of increasing the time between your current and next cast.
Shadow Priests, while we suffer from our main spell being channeled and having to avoid clipping a DoT (2 in 3.2), at least don't have to deal with much in the way of procs that cause us to change our spell casting priority (SimCraft also has values btw for spell "flight-time" where appropriate and the effects that proc on landing rather than casting).
One idea I've had would be to write a combat log parser for Shadow Priests (on a Dummy/Patchwerk fight) that would just be concerned about looking for the actual latencies experienced during the fight and using that information to maybe give some values that could be plugging into the simulator to give a more accurate representation of how things are.
Rawr is not a simulation, its a model, so things will end up being different between Rawr and SimulationCraft. I use a fixed (Average) latency over the length of the fight. When you ask for a 250ms Lag, that is what your average is.
When it comes down to instants, I agree that Lag is definately lower than Channels. Until Blizzard adds in a /cast [queue] or /cast [nocancel] Channel spells are going to end up doing a lot lower DPS than expected, sadly enough, ShadowPriest DPS relies heavily on Channels.
Rawr - Coder of HolyPriest (Healer) and ShadowPriest (DPS) Modules. Get Your Rawr 2.3.x!
When it comes down to instants, I agree that Lag is definately lower than Channels. Until Blizzard adds in a /cast [queue] or /cast [nocancel] Channel spells are going to end up doing a lot lower DPS than expected, sadly enough, ShadowPriest DPS relies heavily on Channels.
Something I petitioned hard for during TBC and then WotLK beta to be told that this wasn't likely to ever happen due to "flavor" reasons first by Tigole and then by Ghostcrawler iirc.
I'm not sure how there's any particular flavor in having a class where their main spell's effectiveness derives to a fairly large degree by where in the world they happen to live, more so than for other classes.
I wonder if it could be a relevant test to have a player with a Mage and a Priest do some Fireballing/Frostbolting on a targetdummy for 2 minutes, counting how many spells he got off, and then compare to a Priest that flays for similar amount of time. Then figure out how many spells he is supposed to get off as Mage vs Priest and try to figure out just how much DPS we are losing on Channel vs Queue-Castable spells.
Rawr - Coder of HolyPriest (Healer) and ShadowPriest (DPS) Modules. Get Your Rawr 2.3.x!
On a non-latency issue (not having played around with Rawr much), I'm curious as to how you've implemented things like Improved Devouring Plague and the Shadowfiend.
With Improved Devouring Plague are you ignoring any spell power that comes from spirit (Glyph of Shadow and Twisted Faith) as well as 2 piece T8.
And what coefficients are you using for the Shadowfiend etc. I ended up setting up a G15 spamming Shadowfiend and let it run over night to collect as much data as I could at different spellpower levels etc. but obviously it was pretty painful.
I ended up with a DPS spell power coefficient of 37.87% (before Shadowcrawl) with a base damage range of 101 to 211 with the damage able to be blocked but not dodged or parried.
I wonder if it could be a relevant test to have a player with a Mage and a Priest do some Fireballing/Frostbolting on a targetdummy for 2 minutes, counting how many spells he got off, and then compare to a Priest that flays for similar amount of time. Then figure out how many spells he is supposed to get off as Mage vs Priest and try to figure out just how much DPS we are losing on Channel vs Queue-Castable spells.
That could come down a bit to the skill of the Shadow Priest (the Mage could be G15'd heh) but nothing wrong with measuring that too.
Yeah, that could work. Although, don't really need the Mage (as a Frostbolt spamming mage as I mentioned can be G15'd) but I guess if you were curious as to if there was any discernable latency when G15'ing could be handy.
If you're just testing Mind Flay chaining, then just chain Mind Flay on the target for 2 minutes, then see how many ticks Recount says you've done and see how far under 120 it is (assuming 0 haste).
The Imp. DP issue was new to me, although I suspect it will not impact DPS a lot. (Nor cause any significant changes in gear relative value).
Shadowfiend has been using a 181-220 damage range, and Coefficient is Spell Power * 15 seconds over 3.5 seconds, with a 10% penalty over 10 attacks. (38.57%)
This may not be 100% accurate but it matched my results. Shadowfiend should not be blocked after using Shadowcrawl. Shadowfiend does suffer a 6% damage reduction against Boss mobs I believe. (As normal with level based partial resists)
Rawr - Coder of HolyPriest (Healer) and ShadowPriest (DPS) Modules. Get Your Rawr 2.3.x!
07/09/09, 1:12 AM
as much in Ulduar, just puts more pressure on the healers. Save it for when you get jetted in ignis or something for example, which from the WWS it looks like she is doing that.
Similar Threads
