Please upgrade your browser for the best possible experience.

Chrome Firefox Internet Explorer
×

Fixing Shadow Tank Spikiness

First BioWare Post First BioWare Post

Berjiz's Avatar


Berjiz
07.27.2013 , 02:23 PM | #501
Quote: Originally Posted by Kitru View Post
It *used* to be. Now, it'll melt your face off if you're not specifically prepared for it. Thermal Tolerance's CD was reduced so that it gets used *right before"* Force Cloak goes off of CD and it punches right through Resilience so that won't even buy you some time (because, you know, Jesse Sky just wants Resilience to be worthless). Of course, it's interesting that this is probably the *only* fight where Phase Walk is actually an asset since you can instantly break LoS by popping behind a pillar with it. When I did it, it was a bit of a rude awakening as I went from being able to do the things that made it a joke of a fight before (using Resilience and Force Cloak to ignore Thermal Tolerance) and discovering that they did nothing. The standard tanking was a joke like always, but it was annoying as hell to find out that they redid Thermal Tolerance specifically to screw over Shadows.
Let me guess saber reflect works on it?


Quote: Originally Posted by MythicalKnight View Post
You keep going on about probability this, probability that. I'm talking about what I've seen and what I've done. All your probabilities are nothing more than educated guesses. Probability isn't ever going to stand up against the chaos that is a RANDOM number generator.
The "chaos" of a number generator isn't chaos at all, infact isn't not even close to it. As Kitru mentioned before it's unsure if real chaos and therefore true random numbers exist. Most random generators work by inserting numbers(the seed) into a function that converts the numbers into a given range. The functions are made in such a away that the number appear random and that they will pass the appropiate statisical test for the chosen distrubition. The seed used in computers are usually the millisecond of the time and when you want more control the seed can be chosen by the user. Given the same seed the exact same random numbers will be generated.

The range these functions usually give is 0 to 1 since it can be shown that given a set of these numbers you can generate the other distrbutions like normal and such. Most(all i think, are there any exceptions?) distributions in swtor are binomial(aka a series of events X1,X2,X3,...XN either they happen or they don't happen) or uniform(like the dmg range for instance). The math discussed here about the probability of a spike is bionomial if we don't consider the dmg ranges.

TLDR: The numbers from random number generators are chosen so they pass statistical tests, not because they are truly random.

xnightshadex's Avatar


xnightshadex
07.27.2013 , 03:26 PM | #502
Quote: Originally Posted by MythicalKnight View Post
You keep going on about probability this, probability that. I'm talking about what I've seen and what I've done. All your probabilities are nothing more than educated guesses. Probability isn't ever going to stand up against the chaos that is a RANDOM number generator.
Statistical models are used to model every aspect of our lives that has even the smallest iota of randomness. Your statement is patently false.

Quote: Originally Posted by Kitru View Post
Secondly, depending up your opinions on chaos theory, there is either no such thing as a truly random number generator or no such thing as a *virtual* truly random number generator. Computers are incapable of actual random number generation since all they do is run processes to get answers. What computers do to achieve what is *perceived* as a random number is to do calculations that are based upon given variables that change without appreciable ability for people to control. Generally, this means going off of time (microseconds are pretty popular) since, if you use a small enough unit, it's beyond the ability of a person to functionally reproduce. There are a bunch of other randomization constructs, but the basics of it is that computers don't actually generate random numbers: they simply generate the *illusion* of random numbers within an entirely predetermined system.
Quote: Originally Posted by Berjiz View Post
Let me guess saber reflect works on it?

The "chaos" of a number generator isn't chaos at all, infact isn't not even close to it. As Kitru mentioned before it's unsure if real chaos and therefore true random numbers exist. Most random generators work by inserting numbers(the seed) into a function that converts the numbers into a given range. The functions are made in such a away that the number appear random and that they will pass the appropiate statisical test for the chosen distrubition. The seed used in computers are usually the millisecond of the time and when you want more control the seed can be chosen by the user. Given the same seed the exact same random numbers will be generated.

The range these functions usually give is 0 to 1 since it can be shown that given a set of these numbers you can generate the other distrbutions like normal and such. Most(all i think, are there any exceptions?) distributions in swtor are binomial(aka a series of events X1,X2,X3,...XN either they happen or they don't happen) or uniform(like the dmg range for instance). The math discussed here about the probability of a spike is bionomial if we don't consider the dmg ranges.

TLDR: The numbers from random number generators are chosen so they pass statistical tests, not because they are truly random.
I have put together a quick computer lesson for all of you following along at home to explain where randomness comes from. Computers have two 'random' devices. One is the hardware random device and the other is in software.

I do a bit of over-simplification to make my argument shorter. For the adventurous, there is a decent Wiki article on this that is fairly high level: http://en.wikipedia.org/wiki//dev/random

/dev/random vs /dev/urandom

/dev/random is truly random. /dev/urandom is psuedorandom and "random enough". For cryptographic usage where true randomness matters, /dev/random is what is used (for you Java developers, its the difference between SecureRandom and Random). I italicize truly as it is as random as everything else in the universe. I am not here to debate philosophy.

/dev/random is populated based on entropic feedback from peripherals. This is generated by minute fluctionations in signals from your monitor and graphics card, keyboard and mouse activity (and signal variation when idle) among other non-predictable things. /dev/random (or HW random for you non-POSIX types) has a major problem though; it is a blocking device. If there is not enough entropic feedback, applications will block until there is sufficient random data to fullfill the request. Additionally, reads of /dev/random are sequential across all threads and processes. This leads /dev/random to appear 'slow' as only one read request can be done at a time across everything running on your system.

To get around this, most operating systems/programming languages provide a psuedorandom device (in the POSIX world, this /dev/urandom). The advantage of /dev/urandom is that it permits non-blocking IO and is very hard to exhaust (since effectively the kernel will just spout nonsense based on a reusable pool of data into it after every read). The down side is that it is not truly random (it is however, random enough for most uses outside of cryptography and security). It is often an algorithm based on the number of occurances a very small time unit since a fixed point in time (such as the POSIX or DOS epochs). For example, the number of nanoseconds that have elapsed since January 1st, 1970, 00:00:00.000 UTC (the POSIX epoch).

I can guarantee you no game developer would use the HW based /dev/random as it would inject uncontrollable, difficult to predict lag spikes depending on what else is running on their systems. Additionally, servers tend to have serious issues keeping /dev/random full during high utilization of randomization as they have less peripherals than a desktop (unless they have dedicated hardware to generate random data) so entropy accumulates very slowly (I know Sun/Oracle Sparc systems have ways around this, but I doubt SWTOR is backed by the Solaris OS running on Sparc chips. It is far more economical in terms of cost to system performance to use commodity x86/x86_64 based hardware running Linux or Windows).

Additionally, using a psuedorandom number generator allows you to have valid unit testing, as for the same seed, the sequence of random data returned is always the same. For those who don't believe me, I can write you a quick program that you can play with to prove it to yourself (since I doubt you would trust my logs).

So, given how many combat rolls are going on across however many threads and how many application instances on each of their backend hardware platforms, I am going to assume with 98% confidence they are using the psuedorandom device if for no other reason that performance.

Why does this matter?

The psuedorandom number generator (/dev/urandom) is deterministic over sufficiently large datasets or samples of random data (even without knowing the seed). Given no outside influence, this might give one pause when considering the statistical models used by the likes of KBN and dipstick.

If you are someone that believes that this fact has any impact on our ability to create valid statistical models then I have some swampland in Florida to sell you.

Given the number of variables in play, such as the number of attacks (controlled by humans) going on in a given piece of software on a piece of hardware at a fixed point in time, the deterministic behavior of the psuedorandom number generator is statistically eliminated by the number of truly random variables involved in its consumption.

TL;DR: The RNG is as close to truly random as matters for the sake of the statistical models used to model incoming damage to tanks. As such, the statistical models are great tool to determine, given a fight, the statistically most likely outcome. This post makes no argument for or against the accuracy of the assumptions used in the statistical models and simply seeks to eliminate the 'randomness' or the RNG as something that has any bearing on this discussion.
War-Machine // Prophecy of the Five
Isaki (Assassin Tank) Elunari (Carnage Marauder) Kozakura (Lethality Operative)
Seraphilia (Merc Healer) Aelara (Shadow Tank) Katsuki (Combat Sentinel)

Kitru's Avatar


Kitru
07.27.2013 , 08:08 PM | #503
Quote: Originally Posted by xnightshadex View Post
I have put together a quick computer lesson for all of you following along at home to explain where randomness comes from. Computers have two 'random' devices. One is the hardware random device and the other is in software.
That was actually really interesting. I didn't know about the cryptographic random mechanism (I knew there was a secure random, but didn't know the mechanism). Very interesting (even though it was a *complete* departure from the actual conversation, lol).
Walls of Text? I *love* Walls of Text!
My New Class Idea
Shadow Class Rep - Suggest/Review Questions Here
Quote: Originally Posted by Fende View Post
Listen to Kitru. Kitru knows all.

THoK-Zeus's Avatar


THoK-Zeus
07.28.2013 , 09:00 AM | #504
Quote: Originally Posted by Ancaglon View Post
For the last time.

1) YOU are the one that repeatedly accuses people of "lieing". Apologize for these attacks now.

2) The word is "lying". Hence, YOU are the one that needs to "lern to reed" and "rite".
1) I will not apologize for anything, if it's the truth.
2) If you find grammar more important then actual content in posts (my "learn to read" was about his incompetence to understand the difference between kephess spikes and frog miniboss spikes), you are in the wrong forum.

I stopped participating in this discussion after this post:: "Probability isn't ever going to stand up against the chaos that is a RANDOM number generator. ". (I am not feeding any trolls.)

Topic:
Getting killed by spikes is mostly about probabilty of a certain damage over a certain time, when we assume constant healing. If we count burst healing from all the different healer classes in, everything would get far more complex, but for a comparision of all the different tank classes it seems rather unimportant.
The problem the spike simulators in the forums have is that they are not counting specific attacksequences and therefore the probability of a certain amount of damage over a certain amount of time excactly in. As far as i saw kbns simulator is just counting the average probabilty of a certain damage, (e.g. probability of a specific attack times the chance of it being defended/shielded/unmigated), but not the probability of a certain amount of damage over a number of attack chains.
That may work for attacks like terminate (which more or less stands alone), but not for the standard attack chains every gcd an operation boss does.
(Second there are also no defensive cooldowns counted into the calculation as far as i see, just the standard migation. Shadows/Assassin tanks should always cycle with their cooldowns, to prevent all lot spike attacks (resilience for nim thrasher...))

Altheran's Avatar


Altheran
07.28.2013 , 11:36 AM | #505
Quote: Originally Posted by THoK-Zeus View Post
(Second there are also no defensive cooldowns counted into the calculation as far as i see, just the standard migation. Shadows/Assassin tanks should always cycle with their cooldowns, to prevent all lot spike attacks (resilience for nim thrasher...))
The problem is that the Shadow's cooldowns are nor more numerous, nor avaliable more often (lower CD), nor more powerful. If we consider that they have to use these CD and do so in order to be as competitive as other tanks in regular situations, without taking into consideration feasibility either, then it puts them behind in their ability to survive critical situation, because it will be inevitable that they have less tools to face them. It would turn the problem into another problem.

THoK-Zeus's Avatar


THoK-Zeus
07.28.2013 , 11:43 AM | #506
Quote: Originally Posted by Altheran View Post
The problem is that the Shadow's cooldowns are nor more numerous, nor avaliable more often (lower CD), nor more powerful. If we consider that they have to use these CD and do so in order to be as competitive as other tanks in regular situations, without taking into consideration feasibility either, then it puts them behind in their ability to survive critical situation, because it will be inevitable that they have less tools to face them. It would turn the problem into another problem.
They are more powerful then the ones pts have and they have the lowest CDs.
For critical situations you have to take them into account to provide a realistic mathematical modell to explore tank spikes. It's not turning any problem into another one as i am not telling anywhere that they "have" to use their defensive cooldowns to survive Spikes.
I was just stating that it's not a realistic modell to not consider them.

Altheran's Avatar


Altheran
07.28.2013 , 02:02 PM | #507
Quote: Originally Posted by THoK-Zeus View Post
They are more powerful then the ones pts have and they have the lowest CDs.
For critical situations you have to take them into account to provide a realistic mathematical modell to explore tank spikes. It's not turning any problem into another one as i am not telling anywhere that they "have" to use their defensive cooldowns to survive Spikes.
I was just stating that it's not a realistic modell to not consider them.
PTs may not have better CDs than Shadows, but Guardians do (all his CDs are similar or superior). So what a Shadow will avoid/reduce, a Guardian will because he will be able to use similar tools at the same time. So what will differentiate them is the damage profile of what get through these similar CDs.
In the end, when comparing a Shadow with a Guardian, including the CD in the mathematical model becomes optional because not counting CDs may be not representative of reality, but it will not change the relative profiles when comparing these two.

nagatamen's Avatar


nagatamen
07.28.2013 , 02:48 PM | #508
Quote: Originally Posted by Kitru View Post
It *used* to be. Now, it'll melt your face off if you're not specifically prepared for it. Thermal Tolerance's CD was reduced so that it gets used *right before"* Force Cloak goes off of CD and it punches right through Resilience so that won't even buy you some time (because, you know, Jesse Sky just wants Resilience to be worthless). Of course, it's interesting that this is probably the *only* fight where Phase Walk is actually an asset since you can instantly break LoS by popping behind a pillar with it. When I did it, it was a bit of a rude awakening as I went from being able to do the things that made it a joke of a fight before (using Resilience and Force Cloak to ignore Thermal Tolerance) and discovering that they did nothing. The standard tanking was a joke like always, but it was annoying as hell to find out that they redid Thermal Tolerance specifically to screw over Shadows.
Quote: Originally Posted by Berjiz View Post
Let me guess saber reflect works on it?
Yes, Saber Reflect works and gives you long enough to get behind the pillar (assuming you don't mess up like I did and prematurely click it). And it even helps with the boost to dps at the same time!

I was fortunate that when I tried to tank Xeno on my shadow that I had a group that could warn me that Phase Walk was the only sure fire way to avoid the damage. It's still irritating that they changed the mechanics in a way that took away some of one tank's strength and gave it to a different tank.

THoK-Zeus's Avatar


THoK-Zeus
07.28.2013 , 02:51 PM | #509
Quote: Originally Posted by Altheran View Post
PTs may not have better CDs than Shadows, but Guardians do (all his CDs are similar or superior). So what a Shadow will avoid/reduce, a Guardian will because he will be able to use similar tools at the same time. So what will differentiate them is the damage profile of what get through these similar CDs.
In the end, when comparing a Shadow with a Guardian, including the CD in the mathematical model becomes optional because not counting CDs may be not representative of reality, but it will not change the relative profiles when comparing these two.
For spike damage you are not comparing a Shadow with a Guardian, you are comparing how a shadow behaves when getting spike damage and how a Guardian behaves when getting spike damage. You can't get defensive cooldowns out of this equation or else the mathematical modell will be incorrect (I hate to say that again).
You can compare them afterwards but you can't make assumptions on one single parameter before you even have a mathematical modell for spike damage.
Also, they have highly different defensive cooldowns, so I really doubt that it's not changing their relative profiles, but even if they had the excact same defensive cooldowns you couldn't tell how they behave on the damage profile a sin tank has (with high shield/abs....) and how they behave on a jugg, cause they are both 2 completely different tanks...

Aieranda's Avatar


Aieranda
07.28.2013 , 04:28 PM | #510
Is the issue that Resilience doesn't effectively "cleanse" the Thermal Tolerance 1 stack debuff after you Force Cloak? I don't remember it doing that on HM last time Xeno was available at 50, but I think it did do that way back when Xeno was brand new. Also, 1 stack isn't going to ruin you. You take +20% damage from 1 stack.

I'm also guessing Saber Reflect is a temporary oversight as well and will be nerfed in a future release to also not allow the mechanic to be specifically cheesed. Then again, 10 stacks is pretty brutal for any tank to take here, so even if it buys it down to less stacks, it's probably not terrible. In that case, it is the Vanguards that should be incensed by the Xeno fight right now, not Shadows and Guardians.

In any case, I'm not surprised. Almost all instances of Resilience behaving like a "cleanse" have been removed from the game. I think they've over-corrected a little bit in this regard, but even so the devs have made their intentions for Resilience clear. They want it to "block" incoming Tech and Force attacks for 3 to 5 seconds as its primary function. In rare cases, they may allow it to also remove "removable" hostile effects as it says, but it's pretty clear they don't mean for this to be something that defeats their boss mechanics, especially those that would normally force a tank swap. I might suggest they drop that little part of "removes hostile removable effects" from the Tool Tip and do away with it completely since the other tanks don't get a cleanse and that seems like the way they want it to behave. However, until they do, I guess trial and error is the way to see if the devs remembered to block out the oft-forgotten "cleanse" effect of Resilience for any given new or updated content piece. Thermal Tolerance is most likely intended to force a tank swap and possibly other creative LoS actions by the team. If Resilience used to bypass that, they obviously are going to nerf it. Most likely they didn't consider it in the original encounter design and now have accounted for it.

Force Cloak did work for me with Xeno on HM.
If your DPS is TOO awesome, you indeed will not be off cool down of Force Cloak before Thermal comes up again, but worst case scenario is that you do a standard tank swap as the mechanic was probably designed to intend.

Ultimately is it unfortunate that Resilience is hit again? Sure. But let's keep our eye on the prize, which is armor/DR and/or the overall spikiness issue. Trying to persuade the devs to make Resilience better isn't going to solve the problems with spikiness, which is the number one issue around here right now.